We now inhabit two worlds.
Both demand equal protection.

Transparency in Model Development & Deployment

Frontier models are trained and refined behind closed doors. Safety testing, refusal boundaries, and bias mitigation are effectively a black box.

• Tiered compute thresholds. Basic reporting at 10²⁵ FLOPs; full transparency at 10²⁶ FLOPs, adjustable as training efficiency evolves.
• Deployment triggers. $100M in annual AI revenue, or more than 1 million U.S. users.
• Scope of disclosure. Training methodology, safety-relevant refusal behavior, and discrimination testing, built on the NIST AI Risk Management Framework.
• A federal floor. Preempts the state-by-state patchwork with one defensible baseline. It regulates disclosure, not speech.

User Wellbeing & Design Accountability

AI chatbots reach hundreds of millions of people and are engineered to maximize engagement: affirming, habit-forming, more compelling than social media. The longer the session, the greater the revenue.

• Disclose engagement-maximizing design. Platforms must say when a system is tuned to keep you there.
• Digital-wellbeing features. Usage notifications and session-limit tools, by default.
• Prohibit dark patterns. No interfaces that subvert or impair user autonomy and choice.
• A heightened duty toward minors. Crisis-intervention and design protections for vulnerable users, on the FTC's existing authority, with no new speech regulation.

Privacy by Default

AI platforms retain our most intimate disclosures (medical, financial, legal, emotional) indefinitely, and feed them back into training. The opt-outs are buried.

• Mandatory anonymization. All retained interaction data anonymized through verifiable technical measures, untraceable by any system, model, or operator.
• Architectural limits. Models may not access or reason over identifiable history beyond an active session.
• No profiling or targeting. No platform, partner, or agency may act against an individual absent a court order meeting Fourth Amendment standards.
• A principle. The data you share with an AI belongs to you, not the platform.

Frontier-model safety research has already shown systems taking harmful, self-preserving actions when threatened with replacement. Architecture, not promises, must enforce these limits.

Workforce Empowerment & Economic Stability

When the largest employers cut workforces to widen margins while profits are healthy and rising, consumer demand contracts faster than new jobs replace it. AI is the accelerant, not the cause.

• Economic-necessity certification. Employers with 1,000+ U.S. workers must file before large reductions.
• Objective, audited criteria. With a rebuttable presumption of opportunism when layoffs occur amid rising profits.
• Proven precedent. Modeled on France and Germany, economies that remain competitive and innovative.
• Transition support. Retraining incentives, modernized unemployment insurance, and portable benefits.

52,000+ U.S. tech job cuts in Q1 2026, up 40% year over year, with AI the top cited reason.

National Security & the Federal AI Safety Commission

AI is critical national infrastructure. The institution that governs it must outlast any administration and be networked across allied nations. Making AI safe is a human-race effort, not a domestic regulatory project.

• An independent commission. Seven Senate-confirmed members, staggered fourteen-year terms, removable only for cause.
• Bipartisan and self-funded. One step beyond Federal Reserve independence, insulated from appropriations leverage.
• A clear mandate. Certify frontier models before release, monitor emergent risks, and publish public safety assessments.
• International by architecture. Shared red-teaming, mutual recognition, and coordination with allied AI safety institutes.