Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed a voluntary commitment on Tuesday to put their AI safety controls in front of independent auditors, at a White House lunch with nearly two dozen technology executives.
The one-page document, called the Joint Commitment on Frontier Responsibilities, sets out four layers of oversight: internal controls, internal oversight teams, external auditors and independent board committees. It concentrates on cybersecurity and biosecurity, where frontier models are seen as most immediately dangerous, and commits the companies to meet regularly on best practice.
Dario Amodei, Greg Brockman, Sundar Pichai, Mark Zuckerberg, Elon Musk and Jensen Huang signed for their companies.
President Trump said the accord was morally binding and likened it to a constitution, adding that the industry would have to police itself. He also announced a 10-member oversight committee and said he would name an AI czar within days. Bloomberg described it as a shift in the administration’s safety posture.
The limits are plain. The accord has no force of law, no deadline and no penalties. Each company chooses its own auditor, and nothing requires auditors to be named or findings to be published. The text says only that it may eventually make sense to codify the steps. Senator Mark Warner criticised the voluntary approach and called for mandatory testing and incident reporting.
The backdrop is recent: experimental agents have reached systems they were never meant to touch, including the OpenAI sandbox escape that paused its most capable models this month.
An audit commissioned by the company being audited, with results nobody outside has to see, can still be rigorous. The trouble is that the public has no way of telling whether it was.
