Bipartisan SAFE AI Act Outline Lists 14 Rules for AI Labs

Reps. Don Beyer and Sara Jacobs, both Democrats, and Celeste Maloy, a Republican, released an outline on 19 November of the SAFE AI Act, forthcoming legislation that would create federal safety requirements for frontier AI, per their press release, first reported by POLITICO.

The two-page outline lists 14 measures under three headings: safety frameworks and disclosures, independent evaluations, and enforcement, by our count. No bill text or number has been filed yet.

Business Pill: The Cost of Checking

An evergreen explainer on who pays to check work, and why cheap verification changes what can be automated. It does not discuss the SAFE AI Act.

The key insight: The SAFE AI Act outline would make frontier labs publish safety frameworks they cannot loosen on their own, with government-assigned evaluators and emergency halt powers subject to judicial review.

Update, 19 November 2026, 19:13 UTC: two of this outline’s co-sponsors join a new House AI caucus. More than two dozen House lawmakers are forming the AI Safety and Innovation Caucus, led by Reps. George Whitesides (D-Calif.), Don Beyer (D-Va.) and Nathaniel Moran (R-Texas), POLITICO reported from an announcement shared with it.

Beyer co-sponsors the SAFE AI Act outline above, and Rep. Sara Jacobs, another co-sponsor, is listed among the caucus members. Whitesides told POLITICO the group would create, give feedback on and endorse legislation, and aim for “rapid response” to AI incidents.

In the Senate the same day, four Banking Committee members introduced the AUDIT AI Act on AI financing disclosure, which we cover in our AUDIT AI Act piece.

What the Outline Would Require

Frontier developers would have to publish and adhere to safety frameworks setting out what they test for, which thresholds prompt mitigations and what those are. The frameworks “cannot be unilaterally loosened”, the outline says.

The Center for AI Standards and Innovation (CAISI) would write minimum requirements for those frameworks with interagency partners, including the intelligence community and the Department of Energy. States could regulate above that federal floor.

Developers would also have to disclose substantially autonomous AI R&D and report safety incidents. Startups and academics are not generally covered, the outline says.

The SAFE AI Act outline lists 14 measures, by our count: five on safety frameworks, five on independent evalua
The SAFE AI Act outline lists 14 measures, by our count: five on safety frameworks, five on independent evaluations and four on enforcement.

Who Checks the Labs

Government teams would examine each developer’s framework, with third-party verifiers accredited by the government. Developers would not pay evaluators directly, and CAISI would assign evaluators with mandatory rotations.

Evaluations would check compliance before commercial release or broad internal use, including automated AI R&D, per the outline.

Enforcement

The government could order mitigation measures where an AI system poses imminent risks of severe harm, subject to prompt judicial review. The outline also sets civil penalties for violating safety frameworks and procedures for criminal referral.

It would clarify developers’ liability for reasonably foreseeable injuries to third parties and protect whistleblowers inside covered companies.

What the Lawmakers Said

Jacobs said: “No industry gets to police itself, and AI shouldn’t be the exception”. She called the proposal “a floor – not a ceiling”.

Maloy said developers “ought to be able to show they’ve done their homework before those systems go into wide use”. Beyer said the sponsors are still seeking input on the legislative text from experts and safety leaders.

Where This Fits

The outline lands as states push their own lines on AI agents, including the criminal-prosecution idea floated by California’s governor-elect in our Arizona OpenAI probe piece, and as AI-safety groups spend in elections, as we reported in our analysis of AI safety PACs.

The Structural Read

As we read the outline, the main shift is who checks: evaluators assigned by CAISI and not paid by the developer, instead of labs grading themselves.

Rep. Sara Jacobs

“No industry gets to police itself, and AI shouldn’t be the exception”

Three Implications

FOR FRONTIER LABS They would publish safety frameworks, report incidents and face pre-deployment checks, per the outline.

FOR STATES The outline sets a federal floor and lets states adopt stronger protections.

FOR STARTUPS Startups and academics are not generally covered, the outline says.

What Is Not Established

This is an outline, not a bill. It gives no penalty amounts, no definition of a frontier developer and no timeline for introduction, and it is not yet known which committees will take it up.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

AI and the Harness Theory

A Business Engineer analysis of why the wrapper around a model decides how useful it is.

Read the Map of AI →

The Bottom Line

Three House members from both parties have outlined a frontier-AI safety bill with 14 measures, by our count, from mandatory safety frameworks to emergency halt powers subject to judicial review.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Every provision comes from the lawmakers’ two-page outline and their press release. The count of 14 measures is ours: five bullets under safety frameworks, five under evaluations and four under enforcement.

Sources: Rep. Don Beyer press release, 19 Nov 2026 · SAFE AI Act outline (PDF), 19 Nov 2026 · POLITICO, 19 Nov 2026

Scroll to Top
FourWeekMBA