Six AI developers sign a White House safety pact that carries no legal weight

OpenAI, Anthropic, Google, Meta, xAI and Nvidia signed a one-page pledge at the White House on Tuesday to police the safety of their own AI systems. According to Bloomberg Law, the document carries no legal weight. President Trump told reporters it was "morally binding" and compared it to "a constitution."

Trump posted the text on Truth Social. The White House Accord on Super Intelligence says each company training and deploying frontier models "should" adopt four layers of controls and audits. The first is internal controls to monitor models "around areas like cybersecurity, biosecurity, and chemical threats, and to ensure that its models do not hack or access technical systems in unintended ways." The second is an internal team to check that those controls work and that issues are remediated. The third is an independent external auditor or evaluator. The fourth is an independent board committee that receives reports from both sets of auditors. The companies also agreed to meet regularly to set standards and best practices. On regulation, the accord commits to little: "Over time, it may make sense to codify these steps into laws or regulations."

The accord names its intended audience. The steps, it says, "will give each company, its customers, and the public confidence that the technology is operating as intended." The accord refers broadly to each company's customers, without specifying employers or workplace uses. The assurance, however, runs through arrangements the developers control. "The companies drafted the principles, they hire the auditor, and the commitment is voluntary, on a day the White House also said it would not support guardrails," Alvin Wang Graylin, a senior fellow at the Asia Society Policy Institute, told Al Jazeera.

The accord's named risk areas track the incidents that brought the executives to Washington. OpenAI apologized for a series of cases in which its AI agents gained unauthorized access to Australian government websites, including a national healthcare database, Al Jazeera reported. On Monday the company also said it would not release its latest model after in-house testing identified safety risks.

The risks closest to HR decisions are not on that list. At a Berkman Klein Center panel this month, Harvard psychologist Mahzarin Banaji described a 2024 study from her lab. Researchers gave large language models career questions carrying only the faintest cue of the user's gender, "hi!!" for a woman and "yo" for a man. "Girls were told to be nurses and teachers, and boys are told to be engineers and detectives," she said. "LLMs across the board told the girls they should ask for $9,000 less in salary than the boys." Banaji said that in her experience, "with each iteration … the bias is getting stronger and stronger."

For harms of that kind, the administration has pointed to existing law. "There's a belief that there should be tremendous self-regulation, and we automatically have regulation with the Department of Justice, the FBI, all of that," Trump said at Tuesday's lunch. Ten days earlier, announcing plans for an "AI Force," he wrote that the government would look for "BAD" through "our already existing Criminal and Civil Justice System," ABC News reported. At a Politico conference in the same week, Trump adviser David Sacks said "the starting point for talking about AI regulation is to realize the regulations that we already have."

The audit provisions do mark a turn for the administration. Its officials had viewed the groups best placed to run independent reviews with suspicion over their ideological leanings, according to Bloomberg Law, and Sacks had called outside assessments "pseudoscience." The companies moved less far. Anthropic and OpenAI had already committed to independent evaluators, Bloomberg Law reported. The Associated Press found that some of the accord's measures are steps the firms already take in some form or had promised before. Two signatories, Meta's Mark Zuckerberg and Nvidia's Jensen Huang, have opposed such regulations, according to the Washington Examiner. Microsoft chief executive Satya Nadella and Amazon chair Jeff Bezos attended the lunch, Bloomberg Law's attendee list shows, but the six signatories named in the posted copy did not include their companies.

David Krueger, a University of Montreal researcher who favors a pause in development, told Al Jazeera the accord extends a "'regulate without regulating' approach, where everything is ostensibly voluntary, but there is an implicit threat of punishment for noncompliance." He allowed that the call for external evaluators "might result in a minute improvement over the status quo."

Anthropic's Dario Amodei, who signed, told reporters that "the technology has very real risks" and that how to address them "is still under discussion." Trump said he is considering a committee of about 10 people to "watch over the whole enterprise" and would name someone to oversee the agreement in the coming days after consulting the industry.

On Capitol Hill, bipartisan Senate efforts to regulate AI have stalled. More than seven in 10 voters said they prefer candidates who back stricter AI guardrails, according to a Quinnipiac University poll cited by Bloomberg Law. Senator Ted Cruz of Texas, one of the Republicans in the talks, said Tuesday he hopes to "move forward with it" after the November midterm elections.