AI & Technology

Establish a comprehensive federal AI regulatory framework modeled on the EU AI Act’s risk-tiered approach. The United States cannot remain the only major democracy with no binding AI regulation while hosting every major frontier AI company on Earth.

• Four-tier risk framework: Banned uses (social scoring, real-time mass biometric surveillance, subliminal manipulation) are prohibited outright. High-risk applications (hiring, lending, housing, insurance, criminal justice, healthcare) require mandatory pre-deployment bias audits, human review, and transparency. General-purpose AI models above computational thresholds must provide model cards, training data disclosure, and safety testing. Research and open-source development face minimal regulation to preserve innovation.
• Federal AI Regulatory Authority: Create a dedicated federal agency with the technical expertise, enforcement authority, and budget to regulate AI — not a toothless advisory committee. Modeled on the structure that gave the FAA authority over aviation safety and the FDA authority over pharmaceutical safety.
• Public incident database: Establish an AI harm incident database modeled on the FAA’s Aviation Safety Reporting System. Mandatory reporting for high-risk AI failures. Public access to anonymized data. The aviation industry’s safety culture was built on transparent incident reporting — AI needs the same.
• Registration and conformity assessment: All high-risk AI systems must be registered with the federal regulator before deployment. Conformity assessments by accredited third parties. Post-market surveillance with mandatory adverse event reporting.

Any AI system used in employment decisions, tenant screening, credit scoring, insurance underwriting, or criminal justice must undergo independent third-party bias audits. Amazon’s hiring AI, the COMPAS recidivism algorithm, and Facebook’s housing ad system all demonstrated that unaudited AI entrenches discrimination along predictable demographic lines.

• Annual independent third-party bias audits for all high-risk AI systems. Results published publicly. Disparate impact standards apply. Companies cannot grade their own homework.
• Private right of action: Individuals harmed by biased AI decisions can sue. The enforcement mechanism cannot depend solely on an overwhelmed FTC — affected individuals need standing to hold companies accountable directly.
• Joint FTC-EEOC rulemaking establishing audit standards, reporting requirements, and enforcement procedures for AI in hiring, lending, housing, and insurance.
• Explainability requirements: When AI makes a consequential decision about an individual — denying a job, a loan, housing, or parole — the affected person has the right to a plain-language explanation of how the decision was made and a meaningful human appeal process.

AI-generated deepfakes pose an existential threat to election integrity. The 2024 New Hampshire primary robocall proved the threat is not hypothetical. Deepfake technology can produce indistinguishable video of any public figure saying anything. Without federal law, each election cycle will be more compromised than the last.

• Federal ban on AI-generated impersonations of candidates within 90 days of any election. Criminal penalties for deepfake election interference. Platform liability for failure to remove labeled deepfakes after notice.
• Mandatory disclosure labels on all AI-generated media. Machine-readable provenance labels embedded in AI-generated content at the point of creation. Platforms must label AI-generated content in feeds.
• Platform accountability: Social media platforms must disclose when AI systems are used for content recommendation and moderation. Algorithmic amplification of AI-generated election content within 90 days of an election triggers liability.
• Election infrastructure protection: Dedicated federal resources for protecting election infrastructure from AI-enabled cyberattacks. Mandatory security standards for AI systems used in election administration.

AI companies have scraped the entire public internet — text, images, voice, video, creative works — to train models worth billions of dollars. The creators and data subjects whose work built these models received nothing. This is the largest uncompensated appropriation of human creative output in history.

• Affirmative opt-in consent: Your data cannot train AI models without your explicit, informed consent. Not buried in terms of service. Not assumed from public posting. Affirmative opt-in, revocable at any time.
• Compensation framework: When personal data, creative works, or likeness are used in commercial AI training, a compensation framework ensures that creators and data subjects share in the value their contributions generate.
• Right to know: Every person has the right to know whether their data was used in AI training datasets. Companies must maintain auditable records of training data provenance.
• Right to opt out retroactively: If your data was used in training without consent, you have the right to demand its removal from future model iterations. Applies to text, images, voice, likeness, and creative works.

The World Economic Forum projects 85 million jobs displaced by AI by 2025 and 97 million new roles created. The new roles require different skills, exist in different industries, and are located in different geographies than the jobs being eliminated. Without a federal transition program, the displacement falls entirely on workers who had no role in the decisions that automated their jobs.

• Federal AI Workforce Transition Fund: $20 billion over 10 years. Portable retraining accounts that follow workers across jobs and industries. Community college AI skills programs. Employer tax credits for retraining existing workers rather than replacing them.
• Advanced notice requirements: Employers deploying AI that eliminates positions must provide 90-day advance notice. Transition assistance is required, not optional. Workers learn about AI displacement from their employer, not from a locked door.
• Sector-specific transition plans for industries facing the highest automation exposure: customer service, data entry, transportation, retail, and administrative support. Each plan developed with worker input and union participation.
• Community impact assessments: When AI deployment will displace a significant portion of a community’s workforce, the deploying company must fund community transition support — modeled on the community impact provisions in plant closure law.

Six companies control over 90% of frontier AI development. The same companies own the cloud infrastructure AI runs on, build the models, and operate the consumer applications. Vertical integration is nearly complete. Without structural intervention, the AI economy defaults to monopoly.

• Structural separation: Companies that build frontier AI models AND control the cloud infrastructure AND operate the application layer face mandatory structural separation. You cannot be the railroad, the freight company, and the department store simultaneously.
• Interoperability mandates: AI APIs must meet interoperability standards. Proprietary lock-in — where switching AI providers requires rebuilding entire systems — is the digital equivalent of the railroad monopoly tactic of incompatible track gauges.
• Open-model requirements: All AI research produced with federal funding must be published as open-weight models. Publicly funded science produces public goods, not proprietary advantages.
• National AI Research Resource (NAIRR): Fully funded at $500M+/year. Public universities guaranteed access to the compute resources that frontier AI development requires. Open-source AI development protected by statute — corporate lobbying cannot ban the open research that drives scientific progress.

Federal and state governments are deploying AI in benefits determinations, sentencing recommendations, predictive policing, surveillance, and immigration enforcement. An Idaho Medicaid algorithm cut home care hours for disabled residents by up to 42% before being struck down in court. Due process applies to government decisions — including government decisions made by algorithms.

• Mandatory disclosure: Every federal agency must publicly disclose all AI systems used in decision-making that affects individual rights. No secret algorithms determining who gets benefits, who gets surveilled, or who gets deported.
• Human review requirement: Any AI-assisted government decision affecting individual rights — benefits eligibility, sentencing, immigration, law enforcement targeting — must include mandatory human review with the authority to override the algorithm.
• Algorithmic impact assessments: Before any government agency deploys an AI system, it must complete a public algorithmic impact assessment analyzing potential disparate impacts, error rates, and civil liberties implications.
• Annual public audits: All government AI systems subject to annual independent audits with published results. An independent AI oversight office within the federal government audits AI use across all agencies.
• Federal moratorium on AI social scoring: No US government agency may deploy social scoring or mass behavioral surveillance AI. This is not China. Government does not rate citizens.