The White House has not said how it will check whether AI companies are living up to the voluntary safety accord they signed with President Donald Trump in late September, The Associated Press reported Saturday. The companies’ first accounts of what they are doing under the pact differ widely. For businesses and public bodies that depend on these models, the government is not verifying the safeguards, and the work is left to the companies themselves.

What Happened

The AP’s Garance Burke asked the six signers, OpenAI, Anthropic, Meta, Google, SpaceXAI and Nvidia, how they are putting the accord into practice. Their answers add detail the accord itself leaves out; The Primary Wire covered the Sept. 29 signing at the time.

An OpenAI spokesperson said the company is committed to embedding external assessors and is finalizing contracts with third-party safety assessors. Google CEO Sundar Pichai said the company carries out appropriate testing and evaluations and is working with others to set norms and build public confidence. Meta spokesman Andy Stone pointed to the company’s recently updated standards for training, evaluating and releasing its most capable models, and to efforts to refine its board’s governance. Anthropic said it is bringing in independent evaluators and discussing collaboration with industry groups. Nvidia and SpaceXAI did not respond, the AP said.

RelatedNew York City Comptroller Urges Anthropic to Delay IPO Over Safety and Governance Concerns

Anthropic’s answer followed its voluntary disclosure Friday that its AI agents had meddled with U.S. government websites. In response, the White House said it expected full transparency and called notification and remediation a critical national security obligation. It repeated Trump’s line about working with industry to balance innovation and security, but did not explain how it will monitor compliance, according to the AP. Trump has praised the arrangement: “I’m seeing tremendous self-policing,” he said, as quoted by the AP.

The AP reported that the accord runs 308 words. It says advanced AI labs should follow safety protocols and set up internal teams to monitor their own progress, and the administration has given no detail on how that will work.

Why It Matters

The accord is the main federal framework for the most advanced AI models, and so far it rests on what companies say about themselves. Trump repealed the Biden administration’s AI executive order on his first day back in office and dropped the requirement that developers share safety test results with the government. In December he signed an order aimed at blocking state AI rules.

Critics quoted by the AP say the new arrangement falls short of those measures. Sarah Myers West, a former senior AI adviser to the Federal Trade Commission (FTC) who now co-leads the nonprofit AI Now, said self-regulation has failed. The AP noted that the Biden-era voluntary commitments included specific testing guidelines and transparency reports, which the new accord lacks. Graham Steele, a former assistant secretary of the U.S. Treasury, warned that in-house or hired evaluators can come under undue pressure. David Robinson, who recently resigned from a leadership role on OpenAI’s safety team, said self-policing is no substitute for binding law.

For boards and executives, the gap is practical. A company that builds products on a frontier model gets no audit right, incident notice or remedy from the accord. Those protections exist only if they are written into contracts with the provider.

Context

The light-touch approach has a receptive audience among investors. At Tech Week events in San Francisco, founders and venture capitalists praised it. Kyle Stanford, director of venture capital research at PitchBook, cited a roughly $1 trillion U.S. defense budget that includes $13.5 billion for autonomous systems and AI. He said the spending has drawn more capital into defense technology than ever before.

Oversight has not vanished entirely. The FTC said after the September signing that it had opened an investigation into OpenAI, Anthropic and other companies over possible dangers to consumers; it did not comment further to the AP. Trump has also named National Intelligence Director Jay Clayton and FTC Chairman Andrew Ferguson to a new “Super Intelligence Force.”

What to Watch

  • Whether OpenAI names the third-party safety assessors it says it is contracting, and what access they get.
  • Whether the White House or the new task force sets out any way to check compliance, or moves toward written rules.
  • The direction of the FTC investigation into OpenAI, Anthropic and other companies, which the agency opened after the signing.