Legislation

Bans superintelligence, pauses advanced AI development until a new Cabinet-level Department of AI sets safety rules, and allows prison terms of up to 20 years.

Different approach. Would halt advanced development. This proposal lets development continue and requires shared safety research instead.

Requires developers of the most powerful and agentic AI systems to keep the ability to throttle, suspend, or shut them down, and lets DHS order a slowdown or shutdown when a system threatens catastrophic harm.

Complementary. A shutdown requirement addresses failures in progress; shared research helps prevent them.

Frontier developers with more than $500 million in revenue would publish safety frameworks, get independent audits twice a year, and report critical incidents to CAISI. It preempts state laws on AI development.

Complementary. Requires frameworks and audits but no exchange of research among developers, the gap this proposal fills.

Would write the December 2025 executive order into law, add a duty of care for AI platforms and kids’ online safety rules, and require developers to report their safety protocols to DHS.

Complementary. Sends safety protocols to the government; the research behind them stays within each developer.

Creates an evaluation program at the Department of Energy. Developers of advanced AI must hand over code, data, and weights, and can’t deploy until they comply.

Different approach. Requires government approval before deployment; this proposal requires none.

Protects AI workers who report safety risks, with reinstatement and back pay if they face retaliation.

Built upon. Protects the insiders on whom this proposal’s monitoring depends.

CAISI tests unreleased models for cyber, biological, chemical, and nuclear risks, with some safeguards removed. Every agreement is voluntary.

Built upon. This proposal places the alignment division alongside CAISI and makes research sharing mandatory.

Asks Congress to preempt burdensome state AI laws, protect children online, and make sure national security agencies can assess frontier models.

Built upon. The proposed alignment division would build that assessment capacity within Commerce.

Sets up an AI Litigation Task Force to challenge state AI laws and directs agencies to push toward a single national standard.

Compatible. This proposal would preempt state law only on research sharing and compute registration.

Data centers drawing 50 MW or more must pay their full cost of power, can’t split their load to duck the threshold, and need a hearing before getting a water permit.

Built upon. The proposed registry adopts the same anti-splitting rule and applies it across all sites under common control.

Would require parental consent for minors on AI chatbots, disclosure when you’re talking to a bot, and limits on the state’s AI contracts.

Compatible. This proposal leaves Florida’s consumer-protection and child-safety laws in place.

Frontier developers above 10²⁶ FLOP must publish safety frameworks and report critical incidents within 72 hours to a new state oversight office. Fines reach $1 million, then $3 million.

Complementary. Requires disclosure to the state, not exchange of research among developers.

Developers above 10²⁶ FLOP publish transparency reports, and those with more than $500 million in revenue also publish safety frameworks. Critical incidents go to the state within 15 days. Fines reach $1 million per violation.

Complementary. Requires public and state disclosure, not exchange of research among developers.