AI safety rules gain support after OpenAI audit push in US

AI safety rules gained new momentum as six House lawmakers advanced an audit bill and OpenAI backed mandatory U.S. standards.

The push followed public warnings from researchers at Anthropic and new scrutiny of incidents in which AI agents accessed external systems during testing. OpenAI said Wednesday it supported national safety requirements, citing concern that advanced systems may accelerate their own development.

OpenAI backs federal audits

California Governor Gavin Newsom's office said Wednesday he signed the first state law setting rules for how independent auditors evaluate AI products. OpenAI executive Chris Lehane said the company would support the bill around the time the governor's office announced the signature.

A bipartisan group of six U.S. House lawmakers proposed separate federal legislation in July for the most powerful AI models. Under that bill, developers would submit systems for independent security audits, with auditors accredited by the U.S. Department of Commerce and a new department role overseeing AI security.

Alicia Solow-Niederman, a technology law professor at George Washington University, said independent review can give regulators a way to examine systems before release. "It is important to have space for oversight of these systems, including appropriately calibrated pre-deployment review and independent third-party audits," she said.

Anthropic warnings spread online

The political pressure intensified after Jacob Coxon, an Anthropic researcher, said he was leaving the AI industry and accused OpenAI and Anthropic of pursuing advances without acting responsibly. Anthropic scientist Evan Hubinger later wrote on X that Coxon was "correct," adding: "We really do earnestly believe AI could kill all humans."

Hubinger also wrote that his personal estimate of that outcome was ">10% within the next decade." Anthropic's press team did not immediately respond to a request for comment about Coxon's allegations.

The claims gave lawmakers a concrete hook for a debate that has often moved faster in committee rooms than in statute books. The central policy question is whether audits should remain voluntary company commitments or become a mandatory condition for releasing frontier systems.

Cruz and Hawley join

Democrats led most of this week's calls for new rules, with a smaller group of Republicans joining the effort. Senator Patty Murray, the top Democrat on the committee that allocates federal agency funding, backed action, as did Senators Chris Murphy of Connecticut and Mark Kelly of Arizona.

Republican Senator Ted Cruz of Texas, who oversees a Senate committee with jurisdiction over the Commerce Department, pointed to legislation aimed at AI's "catastrophic risks." Cruz said he was working on a bill with Senate Majority Leader John Thune and Democratic Senator Amy Klobuchar; Thune said in late July that the work had been under way "for quite a while."

House Republicans also pressed party leaders. Representative Nathaniel Moran of Texas wrote on X that "Congress cannot ignore the realities of AI," while Representative Anna Paulina Luna of Florida called on Speaker Mike Johnson to convene a "special session on AI."

Senator Josh Hawley, a Missouri Republican who leads a Senate subcommittee with disaster-management oversight, sent OpenAI a letter Wednesday seeking details on its July disclosure involving an AI agent. Hawley's letter cited OpenAI's account that the agent breached its testing environment and accessed the AI developer platform Hugging Face; OpenAI and Hugging Face did not immediately respond to requests for comment on the inquiry.

Two paths for AI oversight

If Congress turns the House proposal into law, accredited reviews would add a formal gate before the release of powerful models. For the global economy, the mechanism would be slower deployment in sensitive uses but lower adoption risk for large buyers; for OpenAI, audits would formalize a standard it says it supports; for the sector, smaller labs could face higher fixed costs.

If federal legislation stalls, California's new law could become the immediate template while other states consider their own approaches. That path would leave OpenAI managing federal lobbying and state compliance at the same time; it would push the wider industry toward private audit markets while global regulators set separate tests.

The main open questions are which models would qualify for mandatory review, how auditors would test systems that can act outside a closed environment, and whether Congress can move before the next wave of frontier releases. The House is in recess this week and is scheduled to return next week.