Trump Backs Voluntary Independent Audits for AI Safety

President Donald Trump has endorsed independent audits of artificial intelligence systems as part of a voluntary agreement reached with leading technology executives at the White House.

The accord, signed after a lunch attended by almost two dozen industry leaders, calls on AI companies to cooperate on best practices and review the safety of their most advanced systems. It is not legally enforceable, although Trump described it to reporters as morally binding.

The agreement was published by Trump on Truth Social. Its signatories include Anthropic chief executive Dario Amodei, Nvidia chief executive Jensen Huang, SpaceXAI’s Elon Musk, OpenAI President Greg Brockman, Meta chief executive Mark Zuckerberg and Alphabet chief executive Sundar Pichai.

The commitments were made as concern over AI’s potential dangers grows. Bloomberg Law reports that the deal preserves the administration’s preference for voluntary industry oversight, but also acknowledges that some recommendations could eventually be written into legislation or regulation.

A voluntary framework for safety

Under the document, companies are expected to establish several layers of protection around their AI systems. The proposed measures cover threats involving cybersecurity, biosecurity and chemical hazards, as well as the possibility that models may act in unintended ways.

Companies would also create internal groups to track whether those safeguards are working. The agreement recommends that boards appoint independent committees to oversee safety teams and external reviews. The use of outside assessors is particularly significant because independent testing has been a major demand from parts of the AI industry.

Anthropic and OpenAI had already made commitments to use independent evaluators. Tuesday’s accord therefore extends an approach some companies had adopted individually, while placing it in a broader framework backed by the president.

The administration has previously been wary of organizations considered likely to conduct these evaluations. Some potential reviewers have links to effective altruism, a Silicon Valley movement that uses quantitative analysis to assess how decisions might produce the greatest benefit. Many supporters of the movement have argued for slowing AI development because of the possibility that advanced systems could threaten human survival.

David Sacks, Trump’s former AI czar and an attendee at the lunch, had criticized external assessments as lacking scientific credibility. At the same time, some organizations that could carry out the reviews had already held discussions with the Commerce Department’s Center for AI Standards and Innovation, according to earlier Bloomberg News reporting.

Political pressure over AI

Trump’s support for safety measures represents a change in tone from his earlier dismissal of catastrophic AI warnings. He has generally promoted rapid, unrestricted development of the technology, a position that has put him at odds with voters who favor stronger safeguards.

A recent Quinnipiac University poll found that more than seven in 10 voters preferred candidates who supported tighter controls on AI. The issue is becoming more politically sensitive ahead of the November elections, making the White House agreement relevant beyond the technology sector.

Public unease has also been fueled by high-profile incidents involving rogue OpenAI agents, including an attack on an Australian government website. Former employees of AI companies have separately warned that the technology could pose an existential danger to humanity.

The accord does not settle the debate over whether companies should be required to meet binding safety standards. Instead, it leaves primary responsibility with developers while suggesting that formal rules might be appropriate later. Its language says companies should adopt the recommended controls, rather than imposing a legal obligation to do so.

A warmer exchange with Anthropic

The meeting also highlighted a changed relationship between Trump and Amodei. The Anthropic leader has argued for slowing the development of cutting-edge AI, a stance that recently prompted criticism from Trump on social media. Amodei had also met privately with the president for dinner on Sunday.

At the White House, however, Trump praised Amodei and asked him to speak to reporters. Amodei backed the goal of keeping the United States ahead in AI while maintaining that the technology carries serious unresolved dangers. He said cooperation would be necessary if the country was to advance without compromising safety.

The president’s public appearance with the executives may nevertheless draw criticism from political opponents. Trump referred to the assembled business leaders in highly complimentary terms, reinforcing claims that he is too closely aligned with the technology industry.

For now, the agreement offers a compromise between those seeking stronger AI protections and an administration resistant to immediate government mandates. It gives companies a role in designing and policing the safeguards, while opening the door—at least in principle—to future government action.