Technology & AI

Prohibit mandatory identity-based age verification

Age verification systems that require government ID, biometrics, or persistent tracking to access legal online content are prohibited. People have the right to read and access information without being identified or tracked.

Prohibit mandatory identity-based age verification systems that require government ID biometric data or persistent tracking to access lawful content

No centralized databases

Companies cannot build centralized databases of people's identities or browsing history just to verify age. Any age-check system must work without storing or sharing personal information.

Age verification systems must not create centralized databases of identity or browsing activity

Allow privacy-preserving alternatives

Privacy-preserving age checks — like on-device verification or anonymous credential systems — are permitted alternatives. These methods can confirm a user is old enough without revealing who they are.

Allow age assurance mechanisms that preserve privacy such as local device verification zero-knowledge proofs or non-identifying credential systems

Data minimization and deletion

Any permitted age verification system must collect the least information possible and delete it immediately after the check. Holding on to identity data longer than necessary is prohibited.

Any permitted age verification system must minimize data collection and prohibit retention beyond immediate verification

Narrow scope requirements

Age verification requirements can only apply to specific, high-risk types of content — not to the internet in general. This prevents age checks from becoming a tool for broad access control.

Age verification requirements must be narrowly scoped to specific high-risk content categories and may not be expanded to general internet access

Ban surveillance proxy use

Neither governments nor private companies may use age verification systems as cover for tracking, censoring, or monitoring people's online behavior.

Governments and private entities may not use age verification systems as a proxy for surveillance censorship or behavior tracking

High-risk AI requires review

AI systems with significant potential to affect people's lives or rights must go through a formal safety and accountability review before they can be used publicly.

High-risk AI systems must be subject to heightened governance review before deployment

No automation without accountability

AI tools used in critical areas like healthcare, housing, employment, or policing must always have a human who is responsible for their decisions. No AI system can operate in these areas without human accountability.

AI systems that affect rights liberty healthcare education employment benefits policing or housing must not operate without meaningful human accountability

Rights to notice and review

When an automated system makes an important decision about you, you have the right to know about it, get an explanation, appeal it, and have a human review it.

No fully automated decisions in critical domains without a right to notice explanation appeal and human review

Transparency and auditability

When government agencies use AI to make decisions, those systems must be publicly documented, understandable, and open to independent inspection.

AI systems used in public-sector decision-making must be documented transparent and independently auditable

Liability for harms

The companies and people who build and deploy AI systems are legally responsible for harms those systems cause. They cannot avoid liability just because the decision was made by a machine.

Operators and deployers of AI systems must retain legal liability for harms caused by those systems

Testing for bias and safety

AI systems must be tested for problems like bias, safety failures, and potential misuse — both before they launch and after they are in use.

AI systems must be tested for bias safety reliability and misuse risks before and after deployment

Documentation disclosure

Companies operating covered AI systems must publish documentation about those systems — what they do, how they were trained, and what risks they carry — appropriate to how dangerous they are.

Covered AI systems must maintain model cards documentation or equivalent public-facing disclosures appropriate to risk level

No secret law

Governments and agencies cannot use secret AI decision systems to determine your rights, obligations, or access to services. The rules that govern decisions about you must be knowable.

Secret law or undisclosed AI decision systems must not be used to determine rights obligations or access to public services

Disclosure of algorithmic influence

When an AI or algorithm meaningfully affects what you see, whether you qualify for something, or what actions are taken on your account, you must be told that an algorithm is involved.

People must be told when AI or algorithmic systems materially influence ranking eligibility triage moderation or access decisions

Right to contest decisions

If an algorithm makes a decision that seriously harms you, you have the right to challenge it and ask for a review.

People must have the right to contest materially harmful algorithmic decisions

Non-algorithmic alternatives

Where feasible, platforms must offer users a version of their service that is less personalized or not driven by algorithmic recommendations.

Users must have access to non-algorithmic or less personalized alternatives where practical

Ban manipulative optimization

Algorithms designed to manipulate users by targeting psychological weaknesses — like fear, outrage, or addiction — for profit or political influence are banned.

Ban manipulative algorithmic optimization that targets psychological vulnerabilities for profit or political influence

Disclose ranking objectives

Platforms must explain in plain terms what their recommendation and ranking systems are trying to optimize for — for example, engagement, recency, or relevance.

Platforms must disclose core ranking and recommendation objectives in understandable terms

Protected researcher access

Independent researchers must be able to access platform data to study how algorithms affect people, subject to privacy protections.

Independent researchers must have protected access to platform data needed to study algorithmic harms subject to privacy safeguards

Independent auditing requirement

AI systems that have significant potential to affect people must allow independent auditors to inspect them for safety, fairness, reliability, and potential for misuse.

Covered AI systems must support independent auditing for safety bias reliability and misuse

Meaningful redress for harms

If an AI-assisted decision harms you, you have the right to a meaningful explanation of how that decision was made and a path to seek redress.

Affected individuals must have access to meaningful explanation and redress when harmed by AI-assisted decisions

Forensic logging

Organizations using high-risk AI systems must keep detailed records of decisions so those decisions can be reviewed, investigated, or appealed later.

Organizations deploying high-risk AI must keep logs sufficient for forensic review investigation and appeals

Public AI registry

The government must maintain a public list of all AI systems it uses, with exceptions only for narrowly classified programs — and even those must be overseen by independent reviewers.

Government use of AI must be cataloged in a public registry except for narrowly defined classified exceptions subject to independent oversight

Oversight of classified systems

Even AI systems used in classified government settings must be independently reviewed by authorized oversight bodies. Being classified is not an excuse for avoiding accountability.

Classified or sensitive AI systems must still be subject to independent review under secure oversight procedures

Strict safeguards for high-risk systems

Automated systems with a high risk of serious harm — such as restricting liberty or denying rights — must not be deployed unless strict protections are in place to prevent and address failures.

High-risk autonomous systems must not be deployed where failure could cause loss of liberty bodily harm or deprivation of rights without strict safeguards

No replacement of accountable humans

AI cannot replace the accountable human decision-maker in areas like healthcare, education, housing benefits, or employment. A human must always be responsible.

AI systems used in healthcare education benefits housing or employment decisions must not replace accountable human decision-makers

Prohibit automated denial systems

Fully automated systems that deny people benefits, care, housing, or legal status are prohibited unless human review and the right to appeal are guaranteed before any harm can occur.

Automated denial systems for benefits care housing or legal status must be prohibited unless human review and appeal rights are guaranteed before harm occurs

Human override authority

Government agencies must always retain the ability for a human to override or escalate any automated decision. There must be clear procedures for when and how humans step in.

Government agencies must maintain human override authority and documented escalation procedures for automated systems

Ban mass facial recognition

Police and government agencies cannot use facial recognition technology to scan crowds or public spaces for general law enforcement purposes. This prevents widespread surveillance of people who have done nothing wrong.

Ban mass facial recognition in public spaces for general law-enforcement or administrative use

Ban real-time crowd tracking

Real-time scanning of crowds or protesters using facial recognition or other biometric technology is banned. People have the right to gather and protest without being identified by the government.

Ban real-time biometric tracking of crowds or demonstrations

No general identity infrastructure

Biometric systems — like facial recognition or fingerprint scans — cannot be used as the standard way to verify identity when accessing legal content or participating in public life.

Biometric systems may not be used as general identity infrastructure for access to lawful content or public life

Strict safeguards where permitted

When biometric technology is permitted for a specific purpose, it must be strictly necessary, proportionate to the goal, open to independent review, and governed by clear rules about when data must be deleted.

Where biometrics are permitted they must require strict necessity proportionality auditability and deletion rules

Heightened data protection

Biometric data — like your face scan, fingerprints, or voice — is among the most sensitive kinds of personal information. It requires stronger consent requirements and stricter limits on how long it can be kept than regular personal data.

Biometric data must be treated as highly sensitive protected data with stronger consent and retention rules than ordinary personal data

Prohibit unauthorized cross-agency surveillance fusion

Government agencies cannot combine surveillance datasets from different sources into a unified system unless it is specifically authorized by law, subject to a judge's oversight, and disclosed publicly in general terms.

Prohibit cross-agency fusion of surveillance datasets except where specifically authorized by statute, subject to judicial oversight, publicly disclosed in general terms, and subject to periodic reauthorization

Ban secret dossiers

Government and private actors cannot use AI to secretly build detailed behavioral profiles on individuals. Building such a profile requires specific legal cause and judicial approval.

Ban creation of secret AI-generated behavioral dossiers on individuals absent individualized legal cause and judicial process

Ban government use of commercial data

Government surveillance systems cannot be fed with bulk location data, biometric data, or behavioral data bought from commercial brokers. This closes a loophole that bypasses constitutional warrant requirements.

Ban use of commercially acquired bulk location biometric or behavioral data in AI surveillance systems

AI must enhance not replace learning

AI tools in schools must help students learn — not replace teachers, undermine critical thinking, or create unfair advantages. Students' access to quality education must not depend on technology.

AI systems in education must enhance learning without replacing human instruction critical thinking or equitable access to education

No replacement of human educators

AI cannot take over the job of human teachers in K–12 or higher education. Direct teaching and mentorship require a human educator who can be held accountable.

AI systems may not replace human educators in primary secondary or higher education contexts where direct instruction is required

Support not substitute

AI systems may assist teachers and students, but they cannot substitute for qualified instruction or mentoring. Support tools are allowed; replacement is not.

AI systems may be used to support educators and students but may not substitute for qualified teaching or mentorship

No sole-basis grading

Grades or evaluation outcomes that affect a student's future — like advancing to the next grade, earning a diploma, or disciplinary action — cannot be based solely on AI decisions. A human must be involved.

AI systems may not be the sole basis for grading or evaluation in high-stakes academic decisions including advancement certification or disciplinary action

Right to human review

If an AI system affects your grade or academic evaluation, you have the right to request a human review of that decision.

Students have the right to human review of AI-influenced grading or evaluation decisions

Clear policies on AI use

Schools must have clear policies on how AI may and may not be used in learning. These policies must support genuine learning and cannot rely on AI detection tools that are known to be unreliable.

Educational institutions must establish clear policies on AI use that promote learning and prevent misuse without relying on unreliable detection systems

Evaluate for bias

AI tools used in education must be tested to ensure they treat all students fairly. Tools that disadvantage students based on race, gender, disability, income, or language background are prohibited.

AI systems used in education must be evaluated for bias and must not disadvantage students based on race gender disability socioeconomic status or language background

Accessibility and equity

Every student must be able to use AI tools in the classroom, regardless of their background or resources. AI cannot make educational inequality worse.

AI tools used in education must be accessible to all students and may not create or reinforce educational inequality

Data minimization

When AI systems collect data about students, they may only collect what is actually needed for educational purposes. Gathering extra data about minors is not allowed.

Student data collected by AI systems must be limited to what is necessary for educational purposes

No commercial use of student data

Data collected from students through AI tools cannot be sold, shared with advertisers, or used to build profiles for non-educational purposes.

Student data may not be sold shared or used for advertising profiling or unrelated commercial purposes

Enhanced protections for minors

Children using AI in schools have stronger privacy protections than adults. Stricter rules apply to how student data is collected, stored, and used.

Enhanced privacy protections are required for minors using AI systems in educational settings

Ban invasive surveillance

Schools cannot use AI to continuously monitor students, track their behavior, or analyze their bodies through biometric surveillance without a strong specific justification and protective safeguards.

Ban use of AI systems for invasive surveillance of students including continuous monitoring behavioral tracking or biometric analysis without strict justification and safeguards

No emotion inference

AI systems cannot be used to infer how students are feeling, how attentive they are, or their psychological state as part of grading or discipline decisions.

AI systems may not be used to infer student emotions attention or psychological state for evaluation or discipline

No behavioral scoring

Schools cannot use AI to assign behavioral compliance scores to students. Students are not subjects to be ranked by their conformity to automated expectations.

Educational institutions may not use AI systems to assign behavioral or compliance scores to students

Disclose AI-generated content

When educational content is AI-generated, students must be told. The limitations of AI-generated material must also be disclosed so students can evaluate it critically.

AI-generated educational content must be clearly identified and its limitations disclosed

No ideological manipulation

AI systems in education cannot impose a particular political or ideological viewpoint. Any attempt to use AI to shape students' beliefs without transparency or oversight is prohibited.

AI systems may not be used to impose ideological viewpoints or manipulate educational content without transparency and oversight

Support critical thinking

Education technology should strengthen students' ability to think critically and form their own judgments — not encourage passive acceptance of AI-generated answers.

Educational use of AI must support development of critical thinking rather than passive consumption of generated answers

Accessibility for disabilities

AI tools that help students with disabilities — like transcription, translation, or adaptive learning — are encouraged. Accessibility benefits are a legitimate and important use of AI in education.

AI may be used to improve accessibility for students with disabilities including translation transcription and adaptive learning tools

Multilingual support

AI tools that help students learn in multiple languages are encouraged. Reducing language barriers improves equity — as long as the tools are accurate and fair across languages.

AI systems should support multilingual education and reduce language barriers without compromising accuracy or fairness

Disclosure requirement

Schools must tell students, families, and the public when AI is being used in teaching, grading, or administrative decisions.

Educational institutions must disclose use of AI systems in teaching evaluation and administration

Regular independent audits

AI tools used in education must be regularly inspected by independent reviewers for accuracy, fairness, and whether they are actually helping students learn.

AI systems used in education must be subject to regular independent audits for bias accuracy and educational impact

No proprietary opacity

Technology companies that sell AI tools to schools cannot use business secrets as an excuse to avoid accountability or oversight of how their products work.

Vendors providing AI systems to educational institutions may not use proprietary claims to avoid oversight or accountability

Demonstrate educational benefit

Before AI tools are rolled out to classrooms at scale, schools must have evidence that they actually improve learning outcomes. Unproven tools should not be experimented on students without safeguards.

AI systems must demonstrate educational benefit through evidence before widespread deployment in classrooms

Informed consent for testing

Students cannot be used as unwitting test subjects for AI products. If a school wants to use students to test a new AI tool, families must be informed and given the opportunity to consent.

Students may not be used as unwitting test subjects for AI systems without informed consent and appropriate safeguards

Sustainable operations required

AI companies must not pass their environmental costs — like power consumption, water use, or pollution — onto the public or the environment. They must operate within sustainable limits.

AI systems and infrastructure must not externalize environmental costs and must operate within sustainable limits for energy water materials and ecological impact

Carbon neutrality requirement

Large AI systems and data centers must be carbon neutral or carbon negative over their full life cycle. This means accounting for all emissions from building and running these systems, not just electricity use.

Large-scale AI systems and data centers must meet strict carbon neutrality or carbon-negative requirements over their full lifecycle

Energy disclosure

AI companies must publicly disclose how much energy their systems use, where it comes from, and how much pollution they produce. Transparency is required, not optional.

Operators of large AI systems must publicly disclose energy usage sources emissions and efficiency metrics

Renewable energy supply

Large AI data centers must generate or procure their own clean energy, rather than drawing more from the shared electric grid and driving up costs and emissions for everyone else.

High-consumption AI infrastructure must supply or offset its energy usage through dedicated renewable or low-impact sources rather than relying on general grid extraction

Water usage disclosure

AI infrastructure operators must publicly disclose how much water their facilities use — including for cooling — and what impact that has on local water supplies.

AI infrastructure operators must disclose water usage including cooling consumption and local environmental impact

Water resource protection

AI data centers cannot be built or expanded in ways that strain local water supplies, especially in regions already facing drought or water scarcity.

AI systems must not disproportionately strain local water resources particularly in drought-prone or vulnerable regions

Materials sourcing standards

The rare earth metals and other materials used to build AI hardware must be sourced responsibly, with strong environmental and labor protections throughout the supply chain.

Materials used in AI hardware including rare earth elements must be sourced under strict environmental and labor standards

Lifecycle responsibility

AI hardware makers are responsible for the environmental impact of their products from production through disposal — not just during the use phase.

AI hardware producers must be responsible for full lifecycle impacts including manufacturing durability and end-of-life disposal

Durability and repairability

AI hardware must be designed to last, be repairable, and be reusable. Planned obsolescence that generates unnecessary electronic waste is prohibited.

AI-related hardware must meet durability repairability and reuse standards to reduce electronic waste

Recycling programs

Companies must provide responsible recycling and safe disposal programs for AI hardware and infrastructure components when those items reach end of life.

Operators must implement responsible recycling and disposal programs for AI hardware and infrastructure components

Internalize environmental costs

AI companies must pay for their own environmental impacts rather than shifting those costs to local communities, future generations, or ecosystems.

Companies developing or deploying AI systems must internalize environmental costs rather than shifting them to the public or ecosystems

Environmental impact assessments

Large new AI data centers and infrastructure expansions must undergo formal environmental impact reviews before they can be approved and built.

Large-scale AI deployments must undergo environmental impact assessments prior to construction or expansion

Support climate efforts

AI can and should be used to help address climate change — for example, through climate modeling, environmental monitoring, and optimizing clean energy systems — but these uses must be transparent and accountable.

AI should be used to support climate modeling environmental monitoring and mitigation efforts under transparent and accountable frameworks

Resource optimization

AI tools can be used to reduce energy and water waste in operations — but only if those optimizations don't create hidden environmental harms or unfairly shift burdens onto other communities.

AI may be used to optimize energy water and resource usage provided it does not create hidden environmental tradeoffs or inequitable impacts

Grid modernization support

AI systems may help upgrade and modernize the electrical grid and expand renewable energy — but these tools must operate transparently and serve the public interest, not just corporate efficiency goals.

AI systems may support modernization of electrical grids microgrids and renewable energy systems while maintaining transparency and public accountability

No misrepresentation

AI companies cannot misrepresent their environmental performance by cherry-picking data, using unverifiable carbon offsets, or making sustainability claims that don't hold up to scrutiny.

AI companies may not misrepresent environmental impact through selective reporting offsets or unverifiable sustainability claims

Standardized reporting

Environmental reporting by AI companies must follow standardized, verifiable measurements. Inconsistent or self-selected metrics can hide real impacts.

Environmental reporting for AI systems must follow standardized verifiable metrics to prevent manipulation or selective disclosure

Environmental justice protection

AI data centers and infrastructure cannot be systematically placed in lower-income or marginalized communities to avoid scrutiny or accountability for their environmental harms.

AI infrastructure may not disproportionately locate environmental burdens in disadvantaged communities or regions

No global offloading

The environmental costs of building AI hardware — like mining and manufacturing — cannot be pushed onto lower-income countries or regions that lack equivalent environmental protections.

Environmental costs of AI supply chains must not be offloaded to developing regions without equivalent protections and standards

Integration with national policy

Planning for AI infrastructure must be integrated with national strategies for grid modernization and renewable energy investment. AI's energy needs cannot be treated as separate from public energy policy.

AI infrastructure policy must integrate with national energy grid modernization renewable investment and public infrastructure strategies

Noise pollution standards

Data centers must meet noise pollution standards that protect nearby communities. Local residents must have access to monitoring data and the ability to demand mitigation.

Establish standards for noise pollution from AI data centers and other large data centers including monitoring mitigation and local community protections

Water consumption caps

AI data centers that exceed defined water consumption limits must reduce usage, compensate affected communities, or face regulatory penalties.

Large-scale AI data centers must operate within binding water consumption limits based on regional availability and must report actual usage against permitted caps on a publicly accessible basis

Water-stressed region restrictions

AI data centers cannot be built or expanded in water-stressed regions without a rigorous review of the impact on local water availability and community needs.

AI infrastructure requiring significant water cooling may not be sited in regions classified as water-stressed or drought-prone under federal or state water scarcity designations without demonstrated water-neutral mitigation plans and public review

Groundwater and aquifer protection

AI data centers must not endanger local groundwater or underground water reserves. Protection of underground water sources is required.

AI data centers may not draw on groundwater or aquifer sources in ways that degrade water tables reduce flows to downstream users or damage ecosystems without full environmental review, public notice, and binding mitigation commitments

Water recycling and reclamation

AI data centers must use water recycling and reclamation systems where technically feasible. Wasting water that could be recovered is not acceptable.

New large-scale AI data centers must implement water recycling, reclamation, or closed-loop cooling systems to the maximum extent technically feasible and must report recycling rates alongside gross consumption figures

Drought contingency requirements

AI data centers must have drought contingency plans that specify how they will reduce water consumption during water shortages without shifting burdens onto local communities.

Operators of high-water-consumption AI infrastructure must maintain and publicly file drought contingency plans specifying operational curtailment thresholds during declared water emergencies, priority for municipal and agricultural water users, and timelines for transitioning to low-water or air-cooled alternatives

Community water impact review

Before a large AI data center is built or expanded, the local community must have a formal review process to assess the impact on water supply and other shared resources.

Before approving large data center water permits, regulators must conduct a community water impact review assessing effects on residential water access, municipal supply systems, agricultural users, and environmental flows, with mandatory public comment periods and binding mitigation requirements

Fairness and transparency in finance

AI tools used in finance, credit, and insurance must be fair, transparent, and non-discriminatory. Access to essential financial services cannot be undermined by opaque automated systems.

AI systems in finance credit and insurance must not undermine fairness transparency equal access or protection from discrimination in essential economic systems

No automated denial of credit

AI cannot independently deny you a loan, mortgage, or other financial service. A qualified human must review and be accountable for any denial before it takes effect.

AI systems may not independently deny credit loans mortgages refinancing or other essential financial services without direct human review and accountability

Human review before denial

Any decision to deny, restrict, or worsen your access to credit must be made directly by a human reviewer — not delegated to an AI. The human must make an independent judgment.

Any decision that would deny restrict or materially worsen access to credit or lending must be made directly and independently by a qualified human reviewer before harm occurs

No opaque criteria

AI systems that assess your creditworthiness must be able to explain how they reach their conclusions. Opaque or unexplainable criteria that affect financial decisions are not permitted.

AI systems used in underwriting or creditworthiness assessments may not rely on opaque or unexplainable criteria that materially affect outcomes

Anti-discrimination requirement

AI systems in lending and insurance cannot discriminate based on your race, gender, religion, or other protected characteristics — or based on zip code, language, or other stand-ins for those characteristics.

AI systems in finance and insurance must not discriminate based on protected characteristics or proxies for those characteristics including zip code language behavior or network data

Regular audits for bias

AI systems used in lending, underwriting, and insurance must be regularly audited to check for discriminatory outcomes — including patterns that disproportionately harm particular groups.

AI systems used in lending underwriting pricing or claims decisions must be regularly audited for disparate impact and discriminatory outcomes

No automated insurance denial

AI cannot independently deny, restrict, or cancel your insurance coverage. A human must review and be directly accountable for any adverse coverage or claims decision.

AI systems may not independently deny restrict reduce or terminate insurance coverage or claims without direct and independent human review

Independent human judgment required

Human reviewers making insurance decisions cannot simply rubber-stamp what an AI recommends. They must apply their own independent judgment.

Human reviewers in insurance decisions may not rely solely on AI-generated recommendations and must exercise independent judgment

AI for approval not denial

AI tools can be used to help approve insurance claims or coverage more efficiently — but they cannot be the primary reason a claim is denied, reduced, or restricted.

AI systems may be used to assist or expedite approval of insurance claims or coverage but may not be used as the primary basis for denial restriction or reduction

No negative inference from AI

The fact that an AI did not approve a claim cannot be used — even implicitly — as a reason to deny insurance or delay coverage. AI non-approval is not a valid justification for denial.

Absence of AI approval or recommendation may not be used as evidence or implicit justification for denial restriction or delay of insurance claims or coverage

No behavioral surveillance scoring

AI systems in finance and insurance cannot use broad behavioral surveillance or assign social scores to determine your eligibility, pricing, or access. Your digital behavior cannot be used against you without specific legal authority.

AI systems may not use generalized behavioral surveillance or social scoring to determine eligibility pricing or access in finance credit or insurance

Transparent risk scoring

Risk scoring in finance and insurance must use transparent, lawful, and relevant factors. Systems that rely on opaque behavioral inference to score people are prohibited.

Risk scoring in finance and insurance must be based on transparent lawful and relevant factors and may not rely on opaque behavioral inference

No vulnerability profiling

AI systems cannot use detailed profiles of your personal vulnerabilities to charge you higher prices, fees, or unfavorable terms. Exploiting individual circumstances for profit is prohibited.

AI systems may not use individualized vulnerability profiling to impose exploitative pricing rates fees or terms

Fair housing access

AI tools used in mortgage lending and housing finance cannot reproduce patterns of historical discrimination. Fair access to housing-related financial services must be preserved.

AI systems used in mortgage housing finance or rental screening must not undermine fair access to housing or reproduce historical discrimination

Tenant screening limits

Tenant screening and other housing-related AI systems cannot rely on opaque scoring systems or unverifiable data. People's housing access cannot be blocked by black-box algorithms.

Tenant screening and housing-related AI systems may not rely on opaque scoring or unverifiable data that materially affect housing access

Right to explanation

When an AI system meaningfully influences a financial, credit, or insurance decision affecting you, you have the right to a plain-language explanation of how that decision was made.

People have the right to a meaningful explanation of AI-influenced financial credit or insurance decisions affecting them

Right to appeal

When an AI-influenced decision negatively affects your financial, credit, or insurance situation, you have the right to a timely appeal heard by a human reviewer.

People must have access to a timely human appeal process for materially adverse AI-influenced financial credit or insurance decisions

Disclosure requirement

Companies using AI in consequential financial or insurance decisions must tell you when those systems played a meaningful role.

Entities using AI in consequential financial or insurance decisions must disclose when such systems materially influence outcomes

Data minimization

AI systems in finance and insurance can only collect data about you that is actually necessary for a lawful and relevant decision. Collecting more than needed is prohibited.

AI systems in finance and insurance may collect only data that is strictly necessary for lawful and relevant decision-making

No data sales or repurposing

Financial and insurance data collected by AI systems cannot be sold, shared with advertisers, or repurposed for behavioral profiling unrelated to the original decision.

Financial and insurance data used by AI systems may not be sold shared or repurposed for unrelated commercial targeting or behavioral profiling

No secret profile enrichment

Companies cannot secretly enrich your financial or insurance profile by adding data from third-party brokers, social media, or surveillance sources without explicit disclosure and legal authority.

Entities may not secretly enrich financial or insurance profiles with third-party consumer data social graph data or surveillance-derived data without explicit lawful authority and disclosure

No cross-system denial loops

AI systems cannot trap people in self-reinforcing denial loops where a bad credit score blocks insurance, which blocks housing, which blocks benefits — all without independent human review and due process at each step.

AI systems may not be used to create cross-system denial loops where credit insurance housing or public-benefit decisions reinforce each other without independent review and due process

Independent audits

AI systems in finance and insurance must be regularly, independently audited for bias, fairness, legal compliance, and harm to consumers.

AI systems used in finance credit and insurance must be subject to regular independent audits for bias fairness legality and consumer harm

No proprietary opacity

Companies cannot use business secrets or proprietary claims to prevent regulators, courts, or affected people from understanding, challenging, or appealing AI-influenced financial decisions.

Trade secrecy or proprietary claims may not be used to prevent meaningful oversight explanation or appeal in consequential financial or insurance AI systems

Documentation requirement

Companies deploying consequential AI in finance and insurance must maintain detailed records sufficient to allow independent review, auditing, and accountability.

Entities deploying consequential AI systems in finance and insurance must maintain documentation sufficient for review audit and accountability

No exploitation of vulnerability

AI systems cannot be used to identify and exploit people who are financially vulnerable — targeting them with predatory loans, fees, or insurance terms.

AI systems may not be used to identify and exploit financially vulnerable individuals through targeted loans fees insurance terms or sales practices

No impersonation

AI systems in finance and insurance cannot pretend to be human advisors or misrepresent what authority they have or what obligations they carry. You have the right to know you are dealing with an automated system.

AI systems in finance and insurance may not impersonate human advisors or misrepresent their authority capabilities or obligations to consumers

Public-interest standards for essential systems

When financial or insurance systems function as essential gateways to housing, healthcare, or basic economic participation, they must be governed by the highest public-interest standards — not just market incentives.

Where finance credit or insurance systems function as essential gateways to housing healthcare transportation or economic participation they must be governed by heightened public-interest standards

No automatic exclusion from essential systems

AI systems cannot automatically exclude people from financial systems that are necessary for basic participation in society. Access to essential economic infrastructure must be protected.

AI systems must not be allowed to automatically exclude people from essential economic systems needed for basic social participation

Preserve due process and rights

Government and public-service AI systems must uphold due process, equal protection, and transparency. AI cannot be used to undermine people's legal rights or access to services.

AI systems used by government or public-service entities must not undermine due process equal protection transparency accountability or access to rights and services

No automated denial of benefits

Government AI cannot independently cut off, reduce, or deny your access to public benefits, services, or legal status. Only a human official can make that call.

AI systems may not independently deny terminate reduce or suspend access to public benefits services or legal status determinations

Human decision-maker before harm

Any decision by the government to deny, reduce, or delay benefits or services must be made by a qualified human decision-maker — not delegated entirely to an automated system.

Any decision that would deny reduce terminate or delay benefits services or rights must be made directly and independently by a qualified human decision-maker before harm occurs

Independent human judgment required

When humans review government AI decisions, they cannot simply repeat what the AI recommended. They must think independently and exercise real judgment.

Human reviewers may not rely solely on AI-generated recommendations when making public-service or benefits decisions and must exercise independent judgment

AI for approval not denial

Government can use AI to help identify who might be eligible for benefits or to process approvals more quickly — but AI cannot be the primary driver of denials or terminations.

AI systems may be used to help identify likely eligibility or expedite approvals but may not be used as the primary basis for denial reduction or termination of benefits or services

Right to explanation

If a government AI system affects your benefits, services, legal status, or rights, you have the right to a clear explanation of how that decision was made.

Individuals have the right to a meaningful explanation of any AI-influenced government decision affecting benefits services legal status or access to rights

Right to appeal

When a government AI system makes an adverse decision affecting you, you must have a realistic and timely way to appeal that decision to a human decision-maker.

Individuals must have a timely accessible appeal process before a human decision-maker for any materially adverse AI-influenced government decision

Disclosure requirement

Government agencies must openly disclose when AI systems materially shape decisions that affect the public. Hidden use of AI in government decision-making is prohibited.

Government agencies must clearly disclose when AI systems materially influence decisions affecting the public

No behavioral scoring

The government cannot use AI to assign you a generalized risk score or trustworthiness rating to decide your access to public services or rights. Individual scoring for compliance or suspicion is prohibited.

Government may not use AI systems to assign generalized risk trustworthiness fraud or compliance scores to individuals for access to public services or rights

No discriminatory profiling

Government AI systems cannot use your race, religion, gender, or other protected characteristics — or proxies for them — to infer whether you are eligible, suspicious, or dangerous.

Government AI systems may not use protected characteristics or their proxies to infer eligibility suspicion dangerousness or worthiness

No social conformity scoring

The government cannot use AI to score or monitor your behavior, ideological views, or social conformity as a condition of receiving public services or benefits. Social scoring by government is prohibited.

Government may not use AI systems to monitor or score behavioral conformity ideological alignment or social desirability as a condition of access to services or benefits

Access to human representatives

Everyone must be able to speak with a human government representative for matters involving benefits, legal status, healthcare, housing, education, or other essential services. AI-only channels are not sufficient.

People must retain access to human government representatives for matters involving benefits legal status healthcare housing education or essential public services

No forced AI-only channels

The government cannot force you into an AI-only service channel where the absence of a human would make it harder to be treated fairly, understand decisions, or contest them.

Government may not force individuals into AI-only service channels where lack of human access would impair fairness comprehension or ability to contest decisions

Accessibility for vulnerable populations

AI-powered public services must work for people with disabilities, low-income users, elderly people, and those with limited English proficiency. AI cannot create new barriers to accessing services.

AI-enabled public services must be accessible to disabled low-income elderly and limited-English-proficiency users and may not create new access barriers

No automated disability denials

AI systems cannot deny you disability benefits or override medical evidence from a licensed provider without transparent human review and the opportunity to challenge the decision.

AI systems may not be used to deny disability benefits or override licensed medical evidence without transparent human review and due process

No mass fraud sweeps

The government cannot use AI to conduct mass sweeps looking for fraud or suspicion-based enforcement against people receiving benefits. These actions require individualized legal standards and human review.

Government may not use AI systems to conduct mass fraud sweeps or suspicion-based benefits enforcement without individualized legal standards and human review

No automated termination of essential benefits

AI systems cannot automatically terminate housing assistance, food assistance, healthcare coverage, disability support, or income benefits based on a statistical anomaly or behavioral prediction alone.

AI systems may not automatically terminate housing food healthcare disability or income-support benefits based on scoring anomalies or probabilistic inference

No automated immigration decisions

AI cannot independently decide someone's immigration status, order detention or deportation, rule on asylum claims, or determine family reunification. These decisions require a human official.

AI systems may not independently determine immigration status detention deportation asylum credibility or family reunification outcomes

No credibility inference in immigration

The government cannot use AI to infer whether someone applying for immigration or asylum status is telling the truth, presenting a danger, or has hidden intentions. These judgments require human review.

Government may not use AI systems to infer truthfulness credibility intent or dangerousness in immigration or asylum contexts

Human review for immigration decisions

Any government decision about someone's immigration status or detention that was influenced by AI must be reviewed and finalized by a human official subject to due process and judicial oversight.

Any AI-influenced immigration or detention decision must be reviewed and decided by a human official subject to due process and judicial oversight

Vendor accountability

Private companies that supply AI systems to government must follow the same legal, constitutional, and ethical rules as the government itself. Government cannot outsource its obligations to private vendors.

Private vendors and contractors supplying AI systems to government are subject to the same legal constitutional and ethical constraints as the government entities using them

No proprietary opacity

A company's claim that its AI is a proprietary trade secret cannot be used to block oversight, prevent explanation of decisions, or prevent appeals in public-sector AI systems.

Trade secrecy or proprietary claims may not be used to prevent meaningful oversight explanation or appeal for AI systems used in public decision-making

Procurement disclosure

When government buys AI systems, the procurement process must be public, disclosing what the system does, its limitations, who built it, and how it will be overseen.

Government procurement of AI systems must include public disclosure of purpose capabilities limitations oversight structure and vendor conflicts of interest

Regular independent audits

All major government AI systems must be regularly and independently audited for legal compliance, accuracy, bias, and impact on people's rights.

Government AI systems must undergo regular independent audits for legality accuracy bias rights impacts and administrative fairness

Public AI registry

All government AI systems that affect rights, benefits, or legal status must be listed in a publicly accessible registry that describes what each system does and how it is overseen.

All materially consequential government AI systems must be listed in a public registry with clear descriptions of use authority and oversight

Sunset and reauthorization

Government AI systems that affect rights or services must have expiration dates and must be periodically reauthorized through a public process. Indefinite authorization is not acceptable.

Government AI systems affecting rights benefits or legal status must include sunset dates and periodic reauthorization requirements

Explicit legal authority required

Government agencies cannot deploy AI in ways that affect the public without specific legal authority to do so and clear limits on how the system can be used.

Government agencies may not deploy consequential AI systems without explicit legal authority and defined limits on scope and use

No unwitting test subjects

Government cannot use the general public as uninformed test subjects for consequential AI experiments. Any testing must be lawful, transparent, and subject to ethical oversight.

Government may not use the public as unwitting test subjects for consequential AI systems without lawful authority transparency and ethical safeguards

Right to challenge system legality

People have the right not only to challenge a specific AI decision made against them, but also to challenge whether the AI system itself was legally authorized and valid.

Individuals must be able to challenge not only a government AI decision but also the legality and validity of the system used to make it

Ban opaque immigration AI

AI systems used to make immigration, detention, or deportation decisions must not operate as opaque black boxes. These decisions are too serious to be made by systems that cannot be reviewed or challenged.

Ban use of opaque or unreviewable AI systems in immigration enforcement detention or deportation decisions

Ban risk scoring for detention

AI-generated risk scores used to justify holding someone in immigration detention, denying their release, or separating families are banned.

Ban AI risk scoring systems used to justify immigration detention denial of release or family separation

Common carrier treatment

Internet service providers and the core infrastructure of the internet must be treated as neutral carriers — like telephone or postal services. They must carry all traffic equally.

Internet service providers and core digital network infrastructure must be treated as common carriers or equivalent neutral service providers

Non-discrimination requirement

Network neutrality rules mean ISPs cannot favor or block traffic based on what the content is, who created it, where it is going, or for competitive or political reasons.

Common carrier obligations must prohibit discrimination based on content source destination or political or competitive considerations

No blocking or throttling

Internet providers cannot slow down, block, or prioritize traffic to gain a competitive advantage, earn more money from certain content providers, or advance political goals.

Network operators may not prioritize degrade or block traffic for competitive advantage political influence or economic coercion

Narrow technical exceptions

Network neutrality rules can have narrow exceptions for technical operations like managing network congestion, maintaining security, and ensuring reliability.

Net neutrality frameworks must allow narrowly scoped exceptions for technical optimization security reliability and research and development

Transparent exceptions

Any exceptions to network neutrality must be publicly disclosed, independently auditable, and must not give any company or political actor an unfair advantage.

All exceptions to neutrality must be transparent auditable and prohibited from conferring unfair market or political advantage

Protection from rollback

Network neutrality protections must be written into law, not just regulatory rules. Regulations alone can be rolled back by a new administration; legal protections are more durable.

Core neutrality and access protections must be insulated from administrative rollback through regulatory or executive action alone

No AI sentencing

AI cannot be used to determine how long someone goes to prison, whether they get bail, or what punishment they receive in a criminal case.

AI systems may not be used to determine sentencing bail or punishment in criminal justice proceedings

No risk scoring

AI systems cannot assign someone a score predicting how likely they are to commit another crime, how dangerous they are, or how likely they are to show up for court. These scores cannot be used in judicial decisions.

AI systems may not assign risk scores for recidivism dangerousness or likelihood of compliance in judicial decision-making

No jury profiling

AI systems cannot be used to identify or profile potential jurors based on their behavior, psychology, or demographic background.

AI systems may not be used to profile influence or select jurors based on behavioral psychological or demographic inference

Strict admissibility standards

AI-generated or AI-analyzed evidence must meet strict standards for reliability and transparency, and defendants must have a full opportunity to challenge it before it is used in court.

AI-generated or AI-analyzed evidence must meet strict standards for reliability transparency and cross-examination before being admissible in court

Right to examine AI evidence

Defendants and parties in legal proceedings have the right to examine and challenge any AI system used to generate or analyze evidence in their case — including how it works and what data it used.

Defendants and parties must have the right to examine challenge and obtain disclosure of AI systems used in evidentiary processes

AI for wrongful conviction review

AI may be used to help find wrongful convictions, identify errors, and speed up the appeals process for people who may have been unjustly imprisoned.

AI may be used to assist in identifying wrongful convictions inconsistencies or procedural errors and to accelerate appeals and case reviews

Identify systemic bias

AI can be used to identify patterns of systemic bias in sentencing, policing, and prosecution — as a tool for reform and corrective action, not for making decisions about individuals.

AI should be used to identify systemic bias in sentencing policing and prosecution patterns to support corrective action

No AI final determinations

Judges and court staff cannot use AI to make final decisions about facts, the law, credibility of witnesses, or how cases should come out. These decisions must remain human.

Judges and court staff may not rely on AI systems to make final determinations of fact law credibility or case outcome

Clerical use only with disclosure

AI may be used for clerical tasks in courts — like organizing documents and summarizing routine records — but only with clear disclosure and no effect on legal outcomes.

AI may be used to assist with clerical summarization research and document organization only under clear disclosure and human verification requirements

Disclose material AI use

Courts must disclose any meaningful use of AI in drafting opinions, legal analysis, or summarizing records. Undisclosed AI assistance in judicial work is prohibited.

Courts must disclose material use of AI in opinion drafting legal analysis record summarization or case management where it could affect litigants' rights or case outcomes

No unfair docket prioritization

AI systems cannot be used to prioritize which cases, motions, or hearings get scheduled or heard in ways that create unfairness or unequal access to justice.

AI systems may not be used to prioritize dockets motions or hearings in ways that create unfairness bias or unequal access to timely justice

Preserve accountable reasoning

Any AI-assisted judicial workflow must produce a human-authored, accountable, and reviewable chain of reasoning for every consequential decision.

Any AI-assisted judicial workflow must preserve a human-authored accountable and reviewable chain of reasoning for consequential rulings

Evidence authentication

AI-generated or AI-enhanced evidence must come with documented proof of where it came from, how it was created, and how it was handled.

AI-generated or AI-enhanced evidence must be authenticated with documented provenance methodology and chain of custody before admissibility

Technical disclosure for challenge

Parties in a legal case must receive enough technical information about AI-generated evidence to effectively challenge its reliability and limitations.

Parties must receive sufficient technical disclosure to challenge the reliability validity and limitations of AI-generated or AI-analyzed evidence

Scientific validation required

Courts cannot allow AI outputs to be treated as expert-like evidence unless the AI's methods are scientifically valid, independently tested, and subject to challenge.

Courts may not admit AI outputs as expert-like evidence unless their methods are scientifically valid independently testable and subject to adversarial challenge

Synthetic media high-risk

Deepfakes and synthetic media presented as evidence must be treated as high-risk and must meet enhanced authentication requirements before they can be used in court.

Synthetic media evidence must be presumptively treated as high-risk and require enhanced authentication standards before admission

Preserve analysis logs

If AI is used to analyze evidence in a case, all of the data fed into the AI, the assumptions it made, and every step it took to reach its output must be preserved and available for review.

If AI is used to analyze evidence all material logs model assumptions and transformation steps relevant to the output must be preservable and reviewable

Defense access to AI disclosures

If prosecutors or the government use AI in their investigation, charging decisions, or trial preparation, defendants must have access to the same capabilities and data to ensure a fair proceeding.

If prosecutors or the state use AI systems in investigation charging discovery or litigation support defendants must have equal access to the relevant outputs disclosures and technical challenge mechanisms

Defense funding for AI parity

Public defenders must be funded adequately to keep up with AI tools used by prosecutors. An AI advantage for one side but not the other is fundamentally unfair.

Public defenders and defense counsel must be provided funding and access sufficient to avoid AI-driven asymmetry between prosecution and defense

No AI-driven plea pressure

AI cannot be used to pressure people into accepting plea deals by showing them opaque conviction probability scores or predicted punishments. Plea decisions must be made without AI-driven intimidation.

AI systems may not be used to pressure plea deals through opaque case scoring predicted conviction metrics or punishment leverage models

Guard against prosecutorial AI advantage

Courts must prevent situations where prosecutors have access to powerful AI tools that defense lawyers cannot afford or access. Technological disparity in the courtroom undermines equal justice.

Courts must guard against prosecutorial advantage created by proprietary or undisclosed AI tools unavailable to defense

No juror profiling

AI systems cannot be used to profile potential jurors based on their behavior, psychological traits, or demographic characteristics for the purpose of strategic selection.

AI systems may not be used to profile rank influence or strategically shape juror selection based on behavioral psychological or demographic inference

AI reconstructions must be disclosed

AI-generated reconstructions, simulations, or visualizations shown to a jury must be clearly identified as computer-generated demonstrations — not factual recordings of what actually happened.

AI-generated reconstructions simulations or visualizations shown to juries must be clearly disclosed as generated demonstrative material and subject to strict admissibility standards

No AI summaries to juries

Courts cannot provide juries with AI-generated summaries, interpretations, or credibility assessments of testimony or evidence. Juries must evaluate evidence themselves.

Courts may not provide juries with AI-generated summaries interpretations or credibility assessments of testimony or evidence

No distortion of evidentiary record

AI in the courtroom must not distort the actual evidence, create false impressions of certainty, or present outputs as more precise or neutral than they actually are.

Jury-facing use of AI must not distort the evidentiary record or create false impressions of certainty precision or neutrality

Default to prohibition

The default rule in courts and legal proceedings is that AI use is prohibited except where it is specifically and narrowly permitted with full transparency and audit requirements.

AI use in courts and legal proceedings must default to prohibition except where narrowly permitted for transparent auditable non-generative analytical purposes under strict human oversight

Ban AI-generated evidence

AI-generated evidence — including images, video, audio, text, or reconstructions — cannot be used in legal proceedings.

AI-generated or AI-fabricated evidence including images video audio text or reconstructions is not admissible in legal proceedings

Ban AI-enhanced evidence

AI-enhanced evidence that changes the interpretation or meaning of the underlying evidence is not admissible in court.

AI-enhanced evidence that alters interpretation content or meaning of underlying evidence is not admissible

Analytical use only

AI may only be used for analysis on evidence that already exists — not to generate new content — and only where it does not change or reinterpret that evidence.

AI may be used only for analytical purposes on existing evidence where it does not generate new content and does not alter underlying data

No generative AI for evidence

Generative AI systems, including large language models, cannot be used to analyze evidence in legal proceedings.

Generative AI systems including large language models may not be used for evidentiary analysis in legal proceedings

Verifiable methods only

Analytical AI tools that are permitted in court must use verifiable, reproducible methods — such as established statistical techniques — not opaque generative processes.

Permitted analytical systems must use verifiable reproducible methods such as statistical or traditional machine learning models rather than generative systems

Disclose analytical AI use

Any use of AI in evidence analysis must be explicitly disclosed — including the methods used, what data was inputted, what the limitations are, and what the error rate is.

Any use of AI in evidence analysis must be explicitly disclosed including methods data inputs limitations and error rates

No black-box systems

AI systems that cannot be meaningfully explained, tested, or challenged — so-called black boxes — are not admissible for use in evidentiary contexts in court.

Black-box AI systems that cannot be meaningfully explained tested or challenged are not admissible in evidentiary contexts

AI not expert witnesses

AI systems cannot be brought into court as expert witnesses or treated as authoritative sources on factual or legal matters.

AI systems may not be presented as expert witnesses or as authoritative sources in legal proceedings

Human experts accountable

All expert testimony must come from qualified human experts who are personally accountable for their opinions and can be cross-examined.

All expert testimony must be provided by qualified human experts who are accountable for their conclusions

Disclose AI-assisted expert analysis

A human expert who uses AI tools in their analysis must disclose that fact and remains fully responsible for every conclusion they reach, regardless of what the AI suggested.

Human experts using AI-assisted analysis must disclose such use and remain fully responsible for all conclusions

No AI-drafted opinions

Judges and justices cannot use generative AI to write, draft, or materially shape judicial opinions or rulings. Courts must speak in a human voice, with a human responsible.

Judges and justices may not use generative AI systems to draft write or materially shape judicial opinions or rulings

No AI legal reasoning

AI systems cannot substitute for a judge's own reasoning, legal analysis, or interpretation of the law. Judicial thinking must remain a human function.

AI systems may not substitute for judicial reasoning legal analysis or interpretation of law

Limited clerical use

AI may only be used for limited clerical tasks in judicial settings — like document organization or formatting — that have no effect on legal content or outcomes.

AI may be used for limited clerical tasks such as document organization or formatting only when it does not affect legal reasoning or outcomes

No AI reconstructions to juries

AI-generated reconstructions, simulations, or demonstrative materials cannot be shown to juries. Juries must see and hear actual evidence, not AI-produced dramatizations.

AI-generated reconstructions simulations or demonstrative materials may not be shown to juries

No AI summaries of evidence to juries

AI cannot summarize, interpret, or present evidence or witness testimony to a jury. The jury evaluates evidence directly — AI is not an intermediary.

AI systems may not be used to summarize interpret or present evidence or testimony to juries

No AI influence on jurors

AI cannot directly or indirectly influence how jurors perceive, behave, or make decisions. Jury deliberations must be free from AI manipulation.

AI systems may not be used to influence juror perception behavior or decision-making directly or indirectly

Recognize AI risks

Courts must recognize that AI systems can cause confirmation bias, produce false information, and create false certainty — and must treat all AI outputs with appropriate skepticism.

Courts must recognize that AI systems can amplify confirmation bias hallucination and false certainty and must treat AI outputs as inherently high-risk

Elevated scrutiny standards

Any AI-assisted analysis permitted in court must be held to a higher standard of reliability and scrutiny than ordinary evidence — including rigorous adversarial testing.

Any permitted AI-assisted analysis must meet elevated standards of reliability scrutiny and adversarial testing beyond traditional evidence

Right to challenge AI analysis

All parties must have full rights to challenge any AI-assisted analysis used in their case, including the ability to question the methods, assumptions, and outputs of the AI.

All parties must have full rights to challenge any AI-assisted analysis including methodology assumptions and outputs

Reproducibility requirement

AI-assisted analysis used in court must be reproducible — the opposing party must be able to run the same inputs through the same system and get the same results.

AI-assisted analysis must be reproducible by opposing parties using the same inputs and methods

Preserve analysis logs

All AI-assisted evidentiary analysis must include a complete preserved log of what data was put in, what came out, and every step the system took to get there.

All AI-assisted evidentiary analysis must include preserved logs of inputs outputs and transformation steps

Audit access requirement

Courts and opposing parties must have access to all materials needed to fully audit AI-assisted evidentiary analysis used against them.

Courts and opposing parties must have access to all materials necessary to audit AI-assisted analysis

No AI family court determinations

AI cannot be used to make decisions about child custody, visitation rights, parental fitness, or where a child should be placed — without direct human review and a genuine opportunity to contest the outcome.

AI systems may not be used to determine custody visitation parental fitness or family placement outcomes without direct accountable human judicial decision-making

No parental fitness inference

AI systems cannot draw inferences about a parent's fitness, the risk of abuse, or a witness's credibility from behavioral proxies, psychological profiling, or demographic data.

AI systems may not infer parental fitness abuse risk credibility or child welfare outcomes from behavioral proxies psychological profiling or opaque risk models

No AI in family court recommendations

Family courts cannot rely on AI-generated summaries, recommendations, or credibility assessments in cases involving custody or the welfare of children.

Family courts may not rely on AI-generated summaries recommendations or credibility assessments in matters involving custody abuse allegations or child welfare

Disclose family court AI use

Any AI-assisted tool used in family court administration must be fully disclosed and must not have a material effect on outcomes unless a human reviews and is accountable for that effect.

Any AI-assisted tool used in family court administration must be fully disclosed and may not materially affect outcomes without adversarial challenge rights

No automated evictions

AI cannot automate or drive eviction outcomes without direct human judicial review and a real opportunity to challenge the decision.

AI systems may not be used to automate or materially drive eviction outcomes without direct human judicial review and full due process protections

No tenant risk scoring

Courts handling eviction cases cannot rely on AI-generated tenant risk scores, rental behavior predictions, or opaque housing analytics to decide cases.

Courts may not rely on AI-generated tenant risk scores rental-behavior predictions or opaque housing analytics in eviction or housing-access proceedings

Disclose housing AI use

Landlords who use AI-assisted evidence or analytics in court must disclose this to the other party and provide enough information for the tenant to meaningfully challenge it.

Landlords and housing litigants using AI-assisted evidence or analytics in court must disclose such use and provide sufficient information for challenge and review

No eviction acceleration through AI

Housing courts cannot use AI to speed up case processing in ways that reduce meaningful notice, access to hearings, or the ability to be heard by a judge.

Housing courts must not use AI systems to accelerate case throughput in ways that reduce meaningful notice hearing access or tenant defense rights

No AI in administrative adjudication

AI cannot independently decide the outcome of administrative hearings involving government benefits, licensing, immigration, or other official determinations. A human must make the decision.

AI systems may not independently determine outcomes in administrative hearings involving benefits licensing immigration disability housing or employment rights

No opaque scoring in administrative hearings

Administrative proceedings — like hearings for benefits or licenses — cannot rely on opaque AI scoring or recommendation systems to assess credibility, eligibility, or fitness.

Administrative adjudication may not rely on opaque AI scoring or recommendation systems to determine credibility eligibility compliance or sanction outcomes

Disclose AI in administrative proceedings

When AI materially influences an administrative recommendation, record summary, or decision, the people affected have the right to know about it.

Parties in administrative proceedings have the right to know when AI materially influenced a recommendation record summary or proposed outcome

Administrative appeal rights

Government agencies must provide a genuine human review and meaningful appeal process for any adverse decision that was influenced by AI.

Administrative agencies must provide meaningful human review and appeal for any AI-influenced adverse decision

No AI probation/parole decisions

AI cannot determine conditions of probation or parole, decide whether supervision should be intensified, or recommend revocation — through automated behavioral predictions or scoring.

AI systems may not be used to determine probation parole revocation supervision intensity or conditions of release through opaque risk scoring

No proxy-based supervision scoring

AI systems used in probation and parole cannot assign risk scores based on proxies for race, class, disability, geography, or protected activities. These proxies are well-documented pathways to discriminatory outcomes.

AI systems may not assign supervision risk scores based on proxies for race class disability geography or protected activity

Individualized human review required

Probation and parole decisions must be based on individualized human review — not automated through behavioral analytics or scoring systems.

Probation and parole decisions must be based on individualized human review and may not be automated through behavioral prediction systems

AI to identify bias in probation

AI can be used to identify patterns of bias, inconsistency, or unlawful disparity in probation and parole systems — as a tool for reform and accountability, not for making decisions about individuals.

AI may be used to identify patterns of bias inconsistency or unlawful disparity in probation and parole systems under transparent oversight

No AI fines and fees escalation

AI cannot be used to escalate fines, fees, collection actions, or penalties based on predictions about whether someone will pay. Punishing people for predicted behavior rather than actual conduct is prohibited.

AI systems may not be used to escalate fines fees collections or penalties based on predictive payment scoring or behavioral profiling

No payment coercion through AI

Courts and governments cannot use AI to pressure people into paying through automated threat scoring or penalty escalation. Legal consequences must follow actual due process.

Courts and governments may not use AI to pressure payment compliance through opaque threat scoring or automated penalty escalation

No AI debt-to-punishment escalation

AI cannot convert an administrative debt or missed payment into a harsher legal consequence — like a warrant or license suspension — without direct human judicial review.

AI systems may not be used to convert administrative debt or missed payments into harsher legal consequences without direct human review and due process

AI to identify predatory fine patterns

AI can be used to identify patterns of unjust or predatory fine-and-fee practices for review and correction — as a reform tool, under public oversight.

AI may be used to identify unlawful disparities or predatory fine-and-fee patterns for corrective review under public oversight

AI for legal aid access

AI tools can help people navigate legal aid, understand documents, and access legal information — as long as they do not replace licensed attorneys or give legal advice beyond their capability.

AI may be used to assist legal aid intake document navigation and access to information provided it does not replace licensed legal counsel where counsel is required or appropriate

Disclose legal-aid AI limits

AI tools used in legal aid contexts must be upfront about what they can and cannot do, and cannot present themselves as human lawyers.

AI tools used in legal-aid contexts must clearly disclose their limits and may not present themselves as human attorneys or as a substitute for legal representation

Public legal-aid tools must be fair

AI tools for legal aid must be designed to be accessible, easy to understand, and fair — and must not steer people away from asserting their legal rights.

Public legal-aid AI tools must be designed for accessibility clarity and fairness and may not steer users away from asserting rights or seeking counsel

Fund public legal access tools

Government and courts should invest in public-interest AI tools that help people navigate the legal system — provided these tools improve access without automating critical decisions.

Governments and courts should fund public-interest legal access tools that improve navigation and review without automating final legal judgments

AI records systems must be auditable

AI systems used to search, classify, or summarize court records must be auditable and cannot distort which cases are visible or how they are understood.

AI systems used to classify search summarize or prioritize court records must be auditable and may not distort case visibility or public access

No unequal filing access

Court administrative AI systems cannot create unequal access to filing, scheduling, or records based on hidden preferences, priorities, or biases.

Court administrative AI systems may not create unequal access to filing scheduling or record retrieval based on hidden prioritization logic

Preserve record accuracy

Court records systems that use AI must ensure accuracy, transparency, and a quick way for people to correct errors in their records.

Public court records systems using AI must preserve accuracy transparency and the ability to correct errors promptly

Human override for court admin AI

Courts must maintain human oversight and override capability for all AI-assisted administrative systems that affect people who are parties to cases.

Courts must maintain human override and review mechanisms for all AI-assisted administrative systems affecting litigants or public access

AI limited to assistive functions

Where AI is permitted in judicial settings, it must be limited to transparent, non-generative, assistive functions. It cannot author or materially shape legal reasoning or outcomes.

Where AI is permitted in judicial contexts it must be limited to transparent non-generative assistive functions and may not determine rights liabilities punishment credibility or case outcome

No credibility assessment

AI systems cannot be used to assess whether a witness, defendant, or party is telling the truth, deceiving the court, or has particular intentions. Credibility determination is a human function.

AI systems may not be used to assess credibility truthfulness deception or intent of witnesses defendants or parties in legal proceedings

No behavioral inference in court

Behavioral analytics, biometric analysis, and speech-pattern AI tools cannot be used to infer a person's emotional state, honesty, or reliability in legal proceedings.

Behavioral biometric or speech-analysis AI tools may not be used to infer emotional state truthfulness or reliability in courtroom contexts

No AI alteration of police footage

AI cannot be used to selectively reanalyze, enhance, or reinterpret police body camera or surveillance footage in ways that distort its meaning or context.

AI systems may not be used to reinterpret enhance or selectively analyze police body-camera or surveillance footage in ways that alter evidentiary meaning

Preserve full context in evidence

When AI is used to analyze law enforcement evidence, it must preserve the full context and may not highlight some parts while suppressing others in ways that mislead.

AI-assisted analysis of law-enforcement evidence must preserve full context and may not selectively highlight or suppress information to support a narrative

No sole-basis AI translation

AI translation tools cannot be the sole basis for understanding what was said in a legal proceeding. Automated translation alone is not sufficient for legal accuracy.

AI translation or interpretation systems may not be used as the sole basis for legal understanding in court proceedings where rights or outcomes are affected

Human interpreters required

Human-certified interpreters are required for critical legal proceedings. AI translation may only be used as a supplementary tool, not as the primary interpreter.

Human-certified interpreters are required for critical proceedings and AI translation may only be used as a supplementary tool with disclosure

Forensic AI validation required

AI systems used in forensic laboratories — for things like DNA analysis or ballistics — must meet strict scientific validation standards. They cannot operate as unexplainable black boxes.

AI systems used in forensic laboratories must meet strict scientific validation standards and may not operate as opaque or unreviewable systems

Test forensic AI systems

Forensic AI systems must be independently tested for accuracy, bias, and reproducibility before they can be used in any case that might go to court.

Forensic AI systems must be independently tested for accuracy bias and reproducibility before use in any evidentiary pipeline

No AI reconstruction of sealed records

AI cannot be used to reconstruct, infer, or surface records that have been legally sealed, expunged, or otherwise protected from disclosure — even indirectly through data aggregation.

AI systems may not reconstruct infer or surface sealed expunged or legally protected records through data aggregation or inference

Enforce record protections

Using AI to get around legal protections on sealed or expunged records is prohibited and subject to enforcement.

Use of AI to bypass legal protections on records privacy or expungement is prohibited and subject to enforcement

No AI influence campaigns on juries

AI-generated content cannot be used to try to influence jurors, witnesses, or court proceedings outside the courtroom — for example, through targeted social media campaigns during a trial.

AI-generated content may not be used to influence jurors witnesses or proceedings outside the courtroom through targeted media or messaging

Mitigate public AI influence on trials

Courts must be aware of and work to counter the risk of AI-driven public influence campaigns that could compromise the impartiality of juries or the fairness of proceedings.

Courts must recognize and mitigate risks of AI-driven public influence campaigns that could affect jury impartiality or trial fairness

AI version control required

AI systems used in court proceedings must have version control — meaning they must not change their behavior during an active case.

AI systems used in any permitted judicial context must have version control and may not change behavior during active cases without disclosure and review

Revalidate AI changes

If an AI system used in a legal context changes significantly, it must be revalidated before continued use, and those changes may not retroactively affect prior analyses or decisions.

Material changes to AI systems used in legal contexts must trigger revalidation and may not retroactively affect prior analyses without disclosure

No shift of burden or standards

Using AI tools in a legal proceeding cannot shift the burden of proof, lower evidentiary standards, or reduce any party's procedural rights.

Use of AI systems may not shift burden of proof evidentiary standards or procedural rights of any party in legal proceedings

Transparent AI procurement

Courts must use transparent, publicly disclosed procurement processes when acquiring AI systems — including disclosing what was purchased, from whom, and at what cost.

Courts and judicial systems must use transparent procurement processes for any AI system including public disclosure and review before deployment

Independent validation before deployment

AI systems cannot be put into use in judicial settings without independent validation that they are fair, reliable, and legally compliant.

AI systems may not be deployed in judicial contexts without independent validation for fairness reliability and legal compliance

Authority to suspend AI systems

Courts must be able to suspend or prohibit the use of any AI system that proves to be biased, unreliable, or legally inappropriate.

Courts must maintain the authority to suspend or prohibit use of any AI system that demonstrates bias unreliability or legal risk

Party right to request suspension

Parties to a legal case must have the right to request that an AI system being used in their proceeding be suspended if they have credible grounds for questioning its fairness or reliability.

Parties must have the right to request suspension of AI system use in a case where fairness or reliability is in question

Protect worker rights and dignity

AI tools in employment must not undermine workers' rights, dignity, privacy, or fair access to economic opportunity.

AI systems in employment must not undermine worker rights dignity privacy or fair access to economic opportunity

No automated employment decisions

AI systems cannot make fully automated hiring, firing, promotion, or pay decisions without meaningful human review and accountability.

AI systems may not make fully automated hiring firing promotion or compensation decisions without meaningful human review and accountability

No opaque candidate filtering

AI systems cannot filter or rank job candidates using opaque or unexplainable criteria that materially affect who gets a job or an interview.

AI systems may not be used to filter or rank job candidates using opaque or unexplainable criteria that materially affect employment opportunities

Evaluate for bias

AI systems used in hiring and employment must be tested for bias and cannot result in discrimination based on protected characteristics — or stand-ins for those characteristics.

AI systems used in employment must be evaluated and mitigated for bias and may not result in discrimination based on protected characteristics or their proxies

Right to explanation

Job applicants and employees have the right to receive meaningful explanations of AI-influenced employment decisions that affect them.

Applicants and employees have the right to receive meaningful explanations of AI-influenced employment decisions

Ban intrusive monitoring

Continuous, invasive AI monitoring of workers — including biometric tracking, keystroke logging, and real-time behavioral surveillance — is banned except where strictly necessary and proportionate.

Ban use of AI systems for continuous intrusive monitoring of workers including biometric tracking keystroke logging and real-time behavioral surveillance except where strictly necessary and proportionate

No emotion inference

AI systems cannot be used to infer or monitor workers' emotions, mood, engagement, or psychological state for purposes of employment decisions.

Ban use of AI systems to infer or monitor worker emotions mood engagement or psychological state for employment decisions

No outside-hours monitoring

Employers cannot use AI to monitor or draw inferences about worker behavior outside of working hours.

Employers may not use AI systems to monitor or infer worker behavior outside of working hours

No sole-basis discipline or termination

AI systems cannot be the sole basis for disciplinary action, termination, or changes to pay. A human must be responsible for these decisions.

AI systems may not be used as the sole basis for disciplinary actions termination or compensation changes

Transparent productivity scoring

AI-generated productivity or performance scores must be transparent and auditable, and they cannot be used to affect employment without human review.

AI-generated productivity or performance scores must be transparent auditable and may not be used without human review

No coercive productivity targets

AI systems cannot be used to coerce workers into unsafe, unreasonable, or exploitative productivity targets.

AI systems may not be used to coerce workers into unsafe unreasonable or exploitative productivity targets

Right to refuse harmful AI

Workers must be able to refuse the use of AI systems that materially harm their dignity, privacy, or safety without fear of retaliation.

Workers must have the right to refuse use of AI systems that materially harm their dignity privacy or safety without retaliation

Access to human managers

Workers must have access to human managers or decision-makers for any matter affecting their employment status, discipline, or working conditions.

Workers must have access to human managers or decision-makers for matters affecting employment status discipline or working conditions

Data minimization

AI systems used at work may only collect worker data that is strictly necessary for legitimate business purposes.

AI systems may only collect worker data that is strictly necessary for legitimate business purposes

No data sales or repurposing

Worker data collected through AI systems cannot be sold, shared with third parties, or used for purposes unrelated to the employment relationship.

Worker data collected through AI systems may not be sold shared or used for unrelated purposes

Right to access and correct data

Workers have the right to access, review, and correct data collected about them by AI systems used by their employer.

Workers have the right to access review and correct data collected about them by AI systems

Disclosure requirement

Employers must disclose when AI systems are used in hiring, monitoring, evaluation, and employment decision-making processes.

Employers must disclose use of AI systems in hiring monitoring evaluation and decision-making processes

Regular independent audits

AI systems used in employment must be subject to regular independent audits for bias, fairness, and legal compliance.

AI systems used in employment must be subject to regular independent audits for bias fairness and compliance

Documentation requirement

Employers must maintain documentation of AI systems' purpose, data sources, and decision logic sufficient to allow oversight and accountability.

Employers must maintain documentation of AI system design purpose data sources and decision logic sufficient for oversight and accountability

Ban behavior prediction scoring

AI systems that assign risk scores predicting employee behavior — like likelihood of quitting, unionizing, or engaging in protected activity — are banned.

Ban AI systems that assign risk scores predicting employee behavior such as likelihood of quitting unionizing or engaging in protected activity

Ban anti-union AI

Using AI systems to identify, monitor, or suppress union activity, collective bargaining, or labor organizing is banned.

Ban use of AI systems to identify monitor or suppress union activity collective bargaining or labor organizing

Validate personality assessments

Using AI to infer personality traits or psychological characteristics in hiring decisions is banned unless there is strong scientific evidence and proper safeguards in place.

Ban use of AI systems that infer personality traits or psychological characteristics for hiring decisions without strong scientific validation and safeguards

Worker role in AI deployment

Workers or their representatives must have a meaningful role in reviewing and approving the deployment of high-impact AI systems in their workplace.

Workers or their representatives must have a role in reviewing and approving deployment of high-impact AI systems in the workplace

Collective bargaining over AI

When AI systems affect working conditions, their use must be subject to collective bargaining where applicable.

Use of AI systems affecting working conditions must be subject to collective bargaining where applicable

AI assists not replaces clinicians

AI mental health tools can assist therapists and counselors, but they cannot replace licensed clinicians in diagnosing conditions, evaluating crises, or making high-risk treatment decisions.

AI mental-health tools may assist but must not replace qualified clinicians in diagnosis crisis evaluation or high-risk treatment decisions

Prohibit manipulative AI systems

AI systems designed to create emotional dependency — or to manipulate users through simulated companionship — must be restricted or prohibited. These systems can exploit human needs for connection.

AI systems designed to induce emotional dependency, simulate coercive parasocial attachment, or exploit psychological vulnerabilities for commercial or political ends are prohibited

Evaluate for mental health harms

Platforms and AI products must be evaluated for whether they contribute to addiction, compulsive use, social isolation, or self-harm. Mental health harms from technology must be taken seriously.

Platforms and AI systems must be evaluated for mental-health harms including addiction compulsive use isolation and self-harm amplification

Stronger protections for minors

Children and teenagers need stronger protections against AI systems that are engineered to maximize engagement at the expense of their mental health.

Children and adolescents require stronger protections against manipulative AI systems and engagement optimization

Disclosure in sensitive contexts

When you are sharing something sensitive or emotional, you must be clearly informed if you are talking to an AI, not a human. Deceptive AI personas in emotionally sensitive contexts are prohibited.

Users must be clearly informed when they are interacting with an AI system in emotionally sensitive contexts

Meaningful human control required

When the military or intelligence agencies use AI, meaningful human control must be maintained, and the people giving orders must remain legally and ethically accountable for what the systems do.

Use of AI in military and intelligence contexts must preserve meaningful human control accountability and compliance with constitutional and international law

Ban autonomous lethal targeting

AI systems that can independently identify and kill people — with no human making the actual decision to use lethal force — are banned.

Ban AI systems that can independently select and engage targets with lethal force without meaningful human decision-making and accountability

Ban AI force initiation

AI systems that can start or escalate the use of military force on their own, without a human authorizing each action in real time, are banned.

Ban deployment of AI systems that can initiate or escalate use of force without real-time human authorization

Ban nuclear AI

AI cannot be placed in control of nuclear weapons, command-and-control systems, targeting decisions, or launch processes. Nuclear force must remain under strict human control.

Ban use of AI systems in nuclear command control targeting or launch decision processes

No target filtering or ranking

AI cannot be used to narrow down, filter, or prioritize lists of human targets in a way that effectively drives the targeting decision — even if a human nominally approves the final choice.

AI systems may not be used to narrow, filter, or rank potential human targets for lethal action in a manner that materially influences human decision-making

Identified human decision-maker

When lethal force is used with AI systems involved, there must be a specific, named human decision-maker responsible for that decision. Anonymous AI accountability is not acceptable.

All use of lethal force involving AI systems must include a clearly identified human decision-maker responsible for final authorization

Logging and auditability

All AI-assisted military decisions must be logged and auditable — including what data the AI was given, what it recommended, and what the human decided.

All AI-assisted military decisions must be logged and auditable including data inputs model outputs and human decision points

Transparency for human oversight

AI systems used in military decision-making must be transparent enough for the human operator to genuinely understand and evaluate what the system is recommending before acting on it.

AI systems used in military decision-making must provide sufficient transparency to allow meaningful human understanding and oversight

Testing and reliability

Military AI systems must meet strict reliability and testing thresholds before they are deployed. Unproven AI systems cannot be sent into operational environments.

AI systems used in military contexts must meet strict reliability and testing thresholds before deployment

Ban mass intelligence fusion

AI-driven intelligence surveillance systems that collect or analyze data about civilians on a mass scale without legal authority are banned.

Ban large-scale AI-driven intelligence surveillance systems that indiscriminately collect or analyze civilian data without legal authorization and oversight

No profiling for targeting

AI cannot be used to profile individuals or populations as potential military targets based solely on statistical inference, behavioral patterns, or algorithmic prediction.

Ban use of AI to profile individuals or populations for targeting based on probabilistic or behavioral inference alone

Compliance with laws of armed conflict

All military AI systems must comply with the laws of war — including the principles of distinguishing combatants from civilians, proportionality, and military necessity.

All AI military systems must comply with principles of distinction proportionality and necessity under the laws of armed conflict

Clear attribution of responsibility

When a military AI system takes an action, the legal and moral responsibility for that action must be clearly attributable to specific human beings within a defined command chain.

Responsibility for actions taken using AI systems must be clearly attributable to human actors within a defined chain of command

No evasion of accountability

Military AI systems cannot be used to obscure who is responsible for a military action or to help individuals or governments avoid legal accountability for what they did.

Use of AI systems must not be used to obscure responsibility or evade legal accountability for military actions

Congressional authorization required

Deploying a fundamentally new type of AI-enabled military capability must require explicit authorization from Congress — not just executive decision.

Deployment of new classes of AI-enabled military capability must require explicit congressional authorization

Limits on executive authority

The President and executive branch cannot unilaterally expand the use of AI in warfare beyond what Congress has explicitly authorized.

Executive authority may not unilaterally expand the use of AI in warfare beyond clearly defined and authorized limits

Regular reporting to Congress

All AI-enabled military programs must report regularly to Congress and independent oversight bodies on what they are doing and how they are performing.

Regular reporting to Congress and oversight bodies is required for all AI-enabled military programs

Controlled testing required

Military AI systems must be tested in controlled conditions before being used operationally. No new system goes live without prior evaluation.

AI systems intended for military use must undergo controlled testing and evaluation prior to operational deployment

No real-world testing

Using real military operations — with real people and real consequences — as the primary way to test an unproven AI system is banned.

Ban use of real-world military operations as primary testing environments for unproven AI systems

Adversarial testing

Military AI systems must be adversarially tested — meaning experts must try to find and expose every way the system can fail, be misused, or cause unintended harm.

Require adversarial testing and red-teaming of AI systems to identify failure modes misuse risks and unintended escalation scenarios

Contractor accountability

Private contractors who develop AI for military use must follow the same legal and ethical rules as the government itself. The military cannot outsource its obligations to private companies.

Private contractors developing AI for military use are subject to the same legal and ethical constraints as government entities

No outsourcing of authority

The government cannot hand off responsibility for decisions about the use of force to private contractors or to AI systems. These decisions must remain with accountable government officials.

Government may not outsource decision-making authority or accountability for use of force to private entities or AI systems

Export controls

Strong export controls must exist on high-risk military AI systems to prevent hostile nations or governments with poor human rights records from obtaining and using these technologies.

Establish controls on export of high-risk military AI systems to prevent misuse by hostile actors or regimes with poor human-rights records

Defensive AI permitted

AI systems can be used in defensive roles — like missile defense, threat detection, or intelligence gathering — where they do not make decisions about targeting or killing specific people.

AI systems may be used in defensive or non-person-targeting contexts including missile defense threat detection and interception where the system operates against incoming projectiles vehicles or clearly defined non-human targets

No repurposing defensive AI

Defensive AI systems must not be modified or repurposed to identify, select, or target individual people. A tool approved for defense cannot be converted into an offensive targeting tool.

Defensive AI systems must not be repurposed or extended to identify select or target individual persons or groups

AI for analysis not targeting

AI may be used for analyzing battlefield information, classifying threats, and maintaining situational awareness — as long as it does not generate or recommend specific targets for lethal action.

AI may be used for analysis classification and situational awareness provided it does not generate or recommend specific human targets for lethal action

Ban offensive lethal AI

AI systems cannot be used to execute offensive lethal force — including selecting targets, making engagement decisions, or triggering weapons.

Ban use of AI systems in the execution of offensive lethal force including target selection engagement decisions and weapon deployment

Limited guidance use

AI may be used in weapon guidance systems solely to improve accuracy and reduce civilian casualties — but only after a human has already made the legal decision to engage a specific target.

AI systems may be used in limited weapon guidance roles solely to improve accuracy and reduce collateral damage after a human has independently selected and authorized a lawful target

Guidance parameters strictly defined

AI guidance systems must follow the specific target chosen by a human and must not change, reinterpret, expand, or substitute for that target in any way.

Guidance systems must not alter expand reinterpret or substitute the human-selected target and must operate within strictly defined engagement parameters

No new target selection by guidance

AI guidance systems must be incapable of selecting new targets on their own, reprioritizing targets, or initiating additional attacks beyond the specific engagement the human authorized.

Guidance systems must be incapable of selecting new targets reprioritizing targets or initiating additional engagements

Human responsibility for guidance outcomes

Human operators remain fully responsible for all outcomes when AI-assisted weapon guidance is used — including any unintended harm, collateral damage, or errors.

Human operators remain fully responsible for all outcomes of AI-assisted weapon guidance including unintended harm or collateral damage

Defensive systems against non-human threats

AI can be used in purely defensive systems — like intercepting incoming missiles or detecting attacks — where the purpose is protection rather than targeting people.

AI may be used in defensive systems including missile interception threat detection and protection against incoming attacks where the system operates against non-human targets or active threats

Support international treaties

The United States should pursue international agreements to limit and regulate the use of AI in military and intelligence operations, establishing shared global norms.

Promote and pursue international treaties to limit and regulate the use of AI in military and intelligence contexts

Treaty goal: ban autonomous weapons

International agreements should seek to ban fully autonomous lethal weapons — systems that can kill without a human making the decision — and ensure meaningful human control is maintained.

International agreements should seek to prohibit fully autonomous lethal weapons and ensure meaningful human control over use of force

Treaty goal: ban targeting AI

International treaties should work to ban AI systems that can independently select and engage human targets without a human decision.

International treaties should establish bans on AI systems that independently select and engage human targets

Treaty goal: ban nuclear AI

International treaties should restrict the use of AI in nuclear weapons systems, including command-and-control systems and strategic weapons.

International treaties should restrict AI use in nuclear command control and strategic weapons systems

Treaty mechanisms

International agreements should include transparency requirements and verification mechanisms to confirm that countries are actually complying with military AI restrictions.

Treaties should include transparency reporting requirements and verification mechanisms for high-risk military AI systems

Prevent AI arms races

International efforts should work to prevent a global AI arms race — where countries compete to build the most powerful autonomous weapons — which increases the risk of rapid, accidental escalation.

International efforts should aim to prevent destabilizing AI arms races that increase risk of rapid escalation or accidental conflict

Export controls in treaties

The United States should support international controls on the export of high-risk military AI technologies to prevent these tools from spreading to regimes that would misuse them.

Support international controls on export and proliferation of high-risk military AI technologies to prevent misuse and human rights abuses

Strong sanctions for violations

Countries that violate international agreements on military AI should face strong coordinated sanctions and other consequences from the international community.

Violations of international military-AI treaty obligations by adversaries should trigger strong sanctions and other enforcement measures

Sanctions for prohibited systems

Sanctions for deploying or exporting banned military AI systems should be severe and, where possible, coordinated with allied nations for maximum effect.

Sanctions for deployment or export of prohibited military AI systems should be severe and coordinated internationally where possible

Narrow reciprocal exceptions

If an adversary deploys prohibited military AI systems, limited temporary exceptions to our own restrictions may be considered — but only through proper legal processes and with full public reporting.

If an adversary deploys prohibited military AI systems, limited reciprocal emergency exceptions may be considered only under narrowly defined conditions and independent oversight

Temporary reciprocal measures

Any exception to military AI prohibitions made in response to an adversary's actions must be temporary, reported to Congress to the fullest extent compatible with national security, and subject to legal review.

Any reciprocal exception to military AI prohibitions must be temporary, publicly reported to the fullest extent compatible with security, and subject to sunset and reauthorization

Core prohibitions remain

The core prohibitions — on AI in nuclear weapons, on fully autonomous lethal targeting, and on unaccountable use of force — must remain in place even if adversaries violate international norms. These are non-negotiable floors.

Core prohibitions on nuclear AI control, fully autonomous lethal targeting of persons, and unaccountable use of force remain in effect even during adversary noncompliance

Autonomous targeting as war crime

Using AI to autonomously select or attack human targets, or to meaningfully remove human accountability from decisions about lethal force, is prohibited regardless of what other countries do.

Use of AI systems to autonomously select or engage human targets or to materially remove human accountability in lethal force decisions constitutes a violation of the laws of armed conflict and should be treated as a war crime

Target generation as unlawful

Using AI to generate or recommend human targets for lethal action — or to substitute AI judgment for human judgment on matters of life and death — is prohibited with no exceptions based on reciprocity.

Deployment of AI systems that generate or recommend human targets for lethal action or that substitute for human judgment in use-of-force decisions should be prohibited under international law and treated as unlawful conduct

Violations of armed conflict principles

Using AI in ways that violate the laws of war — including failing to distinguish between combatants and civilians, or acting disproportionately — is prohibited and constitutes a war crime.

Use of AI systems in ways that violate principles of distinction proportionality or necessity under the laws of armed conflict constitutes a war crime regardless of whether a human operator is nominally involved

Precision guidance exception

Using AI solely to guide a weapon more precisely toward a target that a human already lawfully selected — in order to reduce civilian casualties — does not constitute a prohibited autonomous weapons system.

Use of AI solely for precision guidance to reduce collateral damage after lawful human target selection does not constitute prohibited offensive use provided it does not alter or expand the selected target

Commanders remain liable

Individual soldiers and commanders remain personally legally responsible for illegal outcomes that result from using AI systems. Being told to follow an AI recommendation is not a defense.

Individuals and commanders remain legally responsible for unlawful outcomes resulting from use of AI systems and may not use AI as a defense against liability

Dedicated cyber branch

A dedicated, fully developed military cyber branch must be established and maintained to conduct defensive operations, protect national security, and develop strategic cyber capabilities.

Establish a dedicated and fully developed cyber branch within the military focused on defensive and strategic cyber operations

Cyber defense mission

The military cyber branch is responsible for defending critical infrastructure, military systems, and national cyber assets from attack.

The cyber branch shall be responsible for defense of critical infrastructure military systems and national cyber assets against hostile cyber activity

Offensive cyber limits

Offensive cyber operations — attacking another country's or actor's systems — must be strictly limited, authorized at the highest levels of government, and subject to the same legal and oversight requirements as other military operations.

Offensive cyber operations must be strictly limited authorized at the highest levels and subject to the same legal and oversight frameworks governing use of force

Minimize civilian cyber impact

Cyber operations must be designed to minimize harm to civilian infrastructure — including hospitals, utilities, financial systems, and communications — even when targeting adversary systems.

Cyber operations must minimize impact on civilian infrastructure including healthcare utilities financial systems and communications networks

Cyber branch oversight

All military cyber operations must be subject to oversight, audit, and accountability mechanisms equivalent to those applied to other military activities.

All cyber military operations are subject to oversight audit and accountability mechanisms equivalent to other military branches

No domestic cyber surveillance

The military cyber branch cannot be used for domestic surveillance, law enforcement, or monitoring of American civilians, except under the most narrowly defined legal circumstances with judicial oversight.

The cyber military branch may not be used for domestic surveillance law enforcement or civilian monitoring except under narrowly defined and legally authorized emergency conditions

Coordination with civilian agencies

The military cyber branch must coordinate with civilian agencies and private infrastructure operators to protect national security while maintaining clear boundaries between military and civilian roles.

The cyber branch must coordinate with civilian agencies and private infrastructure operators while maintaining clear legal boundaries and accountability

International cyber norms

The military cyber branch should operate in alignment with international efforts to establish norms, agreements, and treaties governing responsible behavior in cyberspace.

The cyber branch should operate in alignment with international efforts to establish norms and treaties governing cyber warfare and infrastructure protection

Public AI surveillance registry

All government AI surveillance systems must be publicly registered, disclosing what they are used for, what legal authority they have, where their data comes from, and how they are overseen.

Require public registration and disclosure of all government AI surveillance systems including purpose authority data sources and oversight structure

Regular independent audits

Every authorized government AI surveillance system must be regularly audited by independent reviewers for legal compliance, accuracy, bias, and civil liberties impacts.

Require regular independent audits of all authorized government AI surveillance systems for legality bias accuracy and civil-liberties impacts

Sunset and reauthorization

All government AI surveillance programs must have expiration dates and must be reauthorized through a public process. Indefinite surveillance authority is not acceptable.

All government AI surveillance authorities and systems must have sunset dates and require periodic reauthorization

Right to access without identification

People have the right to access legal websites and online services without being required to show government ID or prove their identity. Anonymous internet access is a protected right.

Individuals have the right to access lawful online content and services without mandatory identity verification

Protect anonymous use

Using the internet anonymously or under a pseudonym (a made-up name) is a protected right. Free expression and privacy require the ability to communicate without being identified.

Anonymous and pseudonymous use of the internet must be protected as a component of free expression and privacy

Ban warrantless mass surveillance

Mass surveillance of the public without warrants or individualized legal cause is banned. Collecting communications, movements, or activities of people who are not suspected of wrongdoing is prohibited.

Ban AI mass public surveillance

AI-powered surveillance systems that scan public spaces or monitor large numbers of people simultaneously are banned, except in extremely narrow, high-threshold emergency conditions.

Ban AI-powered mass surveillance of public spaces except under narrowly defined high-threshold emergency conditions

Ban persistent location tracking

The government cannot use AI to track your location persistently or stitch together your identity across different platforms and data sources without a court order based on individualized suspicion.

Ban persistent location tracking or cross-platform identity stitching by government without individualized judicial authorization

Ban government data purchases

Government agencies cannot get around constitutional warrant requirements by buying surveillance data from commercial data brokers. Purchasing what you cannot legally collect yourself is prohibited.

Government may not purchase commercially collected surveillance data to evade constitutional limits

Strict warrant requirements

Accessing communications or surveillance data requires strict court-approved warrants, minimization procedures limiting what can be collected, and detailed audit logs. These safeguards are mandatory.

Require strict warrants minimization procedures and audit logs for surveillance access

Ban predictive policing

AI-based predictive policing — using algorithms to guess who might commit a crime — is banned when it relies on opaque scoring systems or data tainted by historical enforcement bias.

Ban predictive policing systems that rely on opaque risk scoring or historically biased enforcement data

Ban suspicion scoring

Government agencies cannot deploy AI tools that assign suspicion or risk scores to members of the public as they go about their daily lives.

Ban suspicion scoring systems for public use by government agencies

Automatic expiration

Surveillance authority must automatically expire unless renewed through a transparent public process. Authorities cannot accumulate indefinite surveillance powers through inaction.

Surveillance authorities must expire automatically unless affirmatively renewed under transparent review

Ban network mapping

AI cannot be used to map out a person's political affiliations, social relationships, religious community, or associations without individualized probable cause approved by a judge.

Ban AI systems that map political social religious or associational networks of individuals absent individualized probable cause and judicial approval

Protect journalists and attorneys

AI-enabled surveillance of journalists, their sources, lawyers, or protected communications requires the highest level of judicial review under the narrowest possible circumstances.

Ban AI-enabled surveillance of journalists sources attorneys and protected communications absent the highest level of judicial review under narrowly defined circumstances

Ban deceptive synthetic media

AI-generated media — also called synthetic media — cannot be used to deceive the public about real people or events in ways that cause harm, manipulate elections, or violate individual rights.

Synthetic media must not be used to deceive the public about real people or events in ways that cause harm, manipulate democratic processes, or undermine individual rights

Ban synthetic impersonation

Using AI-generated images, audio, or video to impersonate a real person for fraud, financial gain, or unauthorized access to systems is banned.

Ban the use of AI-generated media to impersonate a real person for fraud, financial gain, or unauthorized access

Ban identity verification bypass

AI-generated fake identities or media cannot be used to bypass identity verification systems — for example, to fraudulently pass a facial recognition check.

Ban use of synthetic media to bypass identity verification systems

Ban non-consensual sexual content

Creating or distributing AI-generated sexually explicit content depicting a real person without their consent is banned. This is a serious harm to personal dignity and safety.

Ban creation and distribution of non-consensual synthetic sexual content depicting real individuals

Ban false harmful depiction

Creating AI-generated content that convincingly portrays a real person doing or saying something false — in a way that could damage their reputation or put them in danger — is banned.

Ban synthetic media that falsely depicts a real person saying or doing something in a way that would reasonably be understood as fact and would harm their reputation or safety

Ban political manipulation

Using AI-generated content in political advertising that puts fabricated words, actions, or statements in a real person's mouth is banned.

Ban use of synthetic media in political advertising or messaging that depicts real individuals engaging in fabricated actions or speech

Ban misleading about officials

Using AI-generated content to spread false information about a public official's health, actions, or status in order to mislead the public is banned.

Ban synthetic media used to materially mislead the public about the health, actions, or status of public officials

Ban coordinated deception

Using coordinated AI-generated media to deceive the public about real events, elections, or public safety situations is banned.

Ban coordinated use of synthetic media to mislead the public about real-world events, elections, or public safety

Disclosure requirement

When media is substantially AI-generated or has been significantly altered to depict realistic human activity, it must be clearly labeled as such.

Require clear disclosure when media is substantially AI-generated or altered to depict realistic human activity

Provenance markers required

AI-generated video and audio must include built-in, tamper-resistant markers that identify the content as AI-generated. These provenance markers must persist through sharing and distribution.

Require AI-generated video and audio to include persistent, tamper-resistant provenance markers

Open-source detection

The tools needed to detect AI-generated content provenance markers must be freely available to the public, using open-source methods. Verification cannot require expensive or proprietary software.

Provenance markers must be detectable using publicly available, open-source, and free tools

Safeguards against removal

Developers of AI content generation tools must implement reasonable safeguards to prevent users from stripping provenance markers off AI-generated content.

Developers of generative media systems must implement reasonable safeguards to prevent removal or circumvention of provenance markers

Allow parody and satire

AI-generated parody, satire, and artistic expression are allowed — as long as the content is clearly labeled or would not reasonably be mistaken for a genuine factual account.

Allow synthetic media for parody, satire, or artistic expression where it is clearly labeled or would not reasonably be interpreted as factual

Allow disclosed journalism use

Journalists and documentary filmmakers may use AI-generated content for legitimate reporting or storytelling — but only if they clearly disclose it and do not use it to mislead about real events.

Allow use of synthetic media in journalism or documentary contexts when clearly disclosed and not used to mislead about real events

Allow consensual use

AI-generated content depicting a real, identifiable person is allowed when that person has given explicit, informed consent to being depicted in that way.

Allow use of synthetic media depicting real individuals with explicit informed consent