Chips & PolicyLOT 261119 · 3 min read

SAFE AI Act outline would require outside testing of frontier AI

House Democrats and a Republican outline federal safety frameworks, outside evaluations and incident reporting for frontier AI labs, while leaving bill text and final thresholds unresolved.

By Tom Whitfield588 words

Key takeaways

  1. A two-page House outline released Thursday would create the first federal safety requirements for frontier AI developers.
  2. Three representatives—two Democrats and one Republican—released the proposal.
  3. Predeployment evaluations would cover commercial releases, broad internal use and AI used for internal research.
  4. CAISI would assign and rotate evaluators funded through fee-based funds rather than direct developer payments.
  5. Enforcement would include civil penalties, criminal referrals, judicial review of emergency orders and whistleblower protections.
Bipartisan House trio outlines the SAFE AI Act, which would require frontier labs to file safety frameworks and pass gov
Bipartisan House trio outlines the SAFE AI Act, which would require frontier labs to file safety frameworks and pass gov — Mohamed_hassan / pixabay

By Tom Whitfield

A two-page House outline released Thursday would create what the lawmakers call the first federal safety requirements for frontier AI developers, including published frameworks and predeployment evaluations. Reps. Sara Jacobs (D-Calif.), Celeste Maloy (R-Utah) and Don Beyer (D-Va.) released the outline for the Safe and Accountable Federal Evaluations of Artificial Intelligence (SAFE AI) Act.

Their proposal would mandate safety frameworks, incident reporting and more transparency under government oversight. The lawmakers have not released bill text. Beyer said they are still seeking expert input and coordinating with the committees of jurisdiction.

Maloy said the bill “brings in an outside set of eyes to check their work, and leaves room for Utah and other states to set safeguards of their own.” Jacobs said no industry should police itself and called the proposal a floor for state protections rather than a ceiling.

What must frontier developers disclose and report?

Each frontier developer would have to file and publish a safety framework. It would state three things:

  • what the developer tests for;
  • the thresholds that trigger mitigations; and
  • the mitigations themselves.

The Center for AI Standards and Innovation (CAISI), the Commerce Department's AI standards center, would write minimum requirements with partners including the intelligence community and the Energy Department. Developers could not loosen their published frameworks unilaterally.

Developers would also disclose AI research and development described as “substantially autonomous.” Startups and academics would generally fall outside the bill's coverage.

Anyone could confidentially report a safety incident to CAISI. Developers and evaluators would have to report incidents and imminent risks of severe harm. In plain terms, safety reporting would extend beyond company personnel.

Who would evaluate frontier systems, and who would pay?

Government teams would review each developer's framework and run predeployment evaluations. Government-accredited third parties would run compliance evaluations before commercial release or broad internal use.

Those predeployment reviews would include AI used for internal research. CAISI would assign and rotate the evaluators.

Fee-based funds would pay for the work rather than direct developer payments, and limits would restrict how much one company could contribute. The payment and rotation rules are designed to separate the tester from the company being tested.

How would violations be handled?

The outline would begin with civil penalties and higher penalties for knowing or willful violations. It would also permit criminal referrals.

Where safeguards were inadequate and a system posed an imminent risk of severe harm, the government could order mitigations. Any such order would face prompt judicial review.

The proposal would also clarify developers' liability for reasonably foreseeable injuries to third parties and provide whistleblower protections. Its federal floor would let states build additional safeguards on top of national rules. In plain terms, the bill combines financial penalties with emergency controls when severe harm appears imminent.

Why does the proposal arrive now?

The outline arrives as states move ahead with their own rules. New York has opened RAISE Act registration for frontier labs, and a California report this week proposed emergency shutdown powers.

It also came on a day when President Trump told APEC leaders that the United States opposes “globalist” regulation of superintelligence. Beyer described the outline as a roadmap and said he hoped it would foster responsible AI innovation.

The overlap between federal and state proposals means labs could face outside evaluations alongside state registration or proposed shutdown powers. For frontier developers, the final thresholds and enforcement terms will determine the compliance burden and whether federal rules ease or multiply state obligations.

mlq.ai (Source)

Filed under

Share this article:

More from Tom Whitfield

Tom Whitfield

Show full bio

Staff writer covering companies, deals and strategy in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,126 articles

Read also

« Previous articleNext article »