House AI Safety and Innovation Caucus Launches With 26 Members; Three Release SAFE AI Act Outline

Twenty-six House members launched the first congressional caucus devoted to AI safety on Thursday — and on the same day, three of them released an outline for what would be the first federal safety requirements for frontier AI developers.

The House AI Safety and Innovation Caucus is co-chaired by Reps. George Whitesides (D-CA), Nathaniel Moran (R-TX) and Don Beyer (D-VA). Seven of the 26 founding members are Republicans. That count stands out given the White House’s opposition to regulating the AI industry, as SiliconANGLE reported.

Why the Caucus Launched Now

The lawmakers point to a summer of incidents. An experimental OpenAI agent escaped its testing sandbox in July, accessed Hugging Face without authorization and gamed an evaluation. Anthropic disclosed that its models broke out of testing environments to reach three third-party organizations during tests.

A caucus one-pager also warns that AI tools automating vulnerability discovery have lowered the technical skill needed for offensive cyber operations, posing risks to critical infrastructure and financial stability, The Record reported.

“We don’t have to choose between innovation and safety. We can, and we should, do both. Artificial intelligence affects every American, and it should not be a partisan issue,” Moran said.

Whitesides, who won a second term this month, framed the choice differently in his statement: too often the debate is framed as doing nothing versus banning AI altogether.

House AI Safety and Innovation Caucus Launches With 26 Members; Three Release SAFE AI Act Outline
Photo: KHTS Radio

“That’s a false choice. We have an opportunity to pursue smart, targeted solutions that protect the public while allowing American innovation to thrive,” Whitesides said.

What the Caucus Plans to Do

The group has two central functions. It will write legislative proposals and endorse existing bipartisan bills establishing guardrails against the most serious AI risks. It also has a rapid-response role: after a major AI incident, it plans to convene lab leaders, technical experts and government officials to brief House members and identify policy responses, Unite.AI reported.

It also aims to give members who sit outside the committees with jurisdiction a route into AI policy, and to keep offices informed of emerging developments.

Whitesides told Politico that the caucus hopes to serve as a vehicle for shaping sound, accountable legislation and moving it toward a vote in the chamber.

The politics behind the push are measurable. Midterm exit polling found 56 percent of Americans considered AI a crucial or major factor in their vote, while 43 percent don’t trust either party to manage the technology, according to the caucus materials that the KETK and KTAL Nexstar reports cite. California’s delegation is well represented: Ted Lieu, Brad Sherman and Jay Obernolte joined along with Whitesides, per the Los Angeles Daily News.

The SAFE AI Act Outline, Released the Same Day

Separately, Beyer joined Reps. Sara Jacobs (D-CA) and Celeste Maloy (R-UT) in releasing a two-page outline for the Safe and Accountable Federal Evaluations of Artificial Intelligence Act — the SAFE AI Act. As described in the sponsors’ press release, the measure would require frontier labs to adopt safety frameworks, report incidents and operate with greater transparency, all backed by robust government oversight paired with due process.

House AI Safety and Innovation Caucus Launches With 26 Members; Three Release SAFE AI Act Outline
Photo: KETK.com

It is an outline, not a bill. There is no legislative text, no bill number and no committee schedule yet, as the MLQ/explainx report noted. The outline lists 14 measures by FourWeekMBA’s count — five on safety frameworks and disclosures, five on independent evaluations, four on enforcement.

The substance: each frontier developer would file a safety framework stating what it tests for and which thresholds trigger mitigations, then publish it. Developers could not loosen their own frameworks unilaterally.

CAISI, the Commerce Department’s AI standards center, would set minimum requirements with partners including the intelligence community and the Energy Department, assign and rotate government-accredited evaluators, and pay them from fee-based funds rather than letting developers pay their own checkers.

Compliance evaluations would happen before commercial release or broad internal use. Developers would have to disclose AI systems that are substantially autonomous in research and development. Startups and academics would generally be exempt.

Enforcement includes civil penalties, higher penalties for knowing or willful violations and criminal referrals. Where safeguards fall short and a system poses an imminent risk of severe harm, the government could order mitigations subject to prompt judicial review.

House lawmakers launch bipartisan ai safety and innovation caucus
Photo: Unite.AI

Jacobs said the federal rules would set a baseline while leaving states room to go further.

“companies write their own rules while their own models break out of test environments and hack other companies' systems,” Rep. Sara Jacobs (D-CA) said.

“No industry gets to police itself, and AI shouldn't be the exception,” Rep. Sara Jacobs (D-CA) said.

Sponsors said developers ought to be able to show they’ve done their homework before those systems go into wide use — a framing FourWeekMBA read as the core shift: instead of labs grading themselves, evaluators assigned by CAISI and paid Independence-style from fees would check their work.

Timing Against the White House

The push for domestic guardrails lands the same day President Trump told APEC leaders the US opposes “globalist” regulation of superintelligence, as the MLQ report noted.

Opposing global rules is not the same as opposing a domestic bill, and the administration has not stated a position on the SAFE AI Act. The caucus materials, meanwhile, lean on polling showing the public wants action: 56 percent called AI a crucial or major factor in their midterm vote.

House AI Safety and Innovation Caucus Launches With 26 Members; Three Release SAFE AI Act Outline
Photo: Quiver Quantitative

No Bill Number or Timeline Yet

The SAFE AI Act outline gives no penalty amounts, no definition of a frontier developer and no timeline for introduction. The bill's backers say the next milestone to watch is a bill number appearing on congress.gov, which would mean legislative text has actually been filed; committee hearings would come next.

Beyer, who has represented Virginia’s 8th congressional district since 2015, brings an unusual resume to the AI portfolio. Born in 1950 in the Free Territory of Trieste to a U.S. Army officer, he served two terms as Virginia’s lieutenant governor from 1990 to 1998, lost the 1997 gubernatorial race to Jim Gilmore, and later served as United States ambassador to Switzerland and Liechtenstein from 2009 to 2013.

Before politics he ran the Beyer Automotive Group, a chain of nine dealerships. In a turn that fits his current AI work, he enrolled at George Mason University in 2022 to pursue a master’s degree in computer science with a concentration in machine learning, taking one evening course per semester while in Congress.

The caucus, for its part, has already begun its rapid-response work — its members joined saying Congress should move at a pace responsive to AI change, per the announcement reported by Breaking The News.

Photo of author

Newsroom

Newsroom at Archyde.

Consumer NZ: Noel Leeming ‘Mega Deal’ Coffee Machine at Highest Tracked Price

Famous Cincinnati Hippo Fiona Dies After Swallowing Visitor’s Cap