Trump, Tech Giants Sign Voluntary AI Self-Policing Accord
US President Donald Trump and the leaders of the country’s most powerful artificial intelligence companies signed a voluntary, non-binding accord at the White House on September 29, committing the firms to internal controls, independent audits, and board-level oversight of their most advanced models. The one-page document — which Trump posted on Truth Social and described as “morally binding” — stopped well short of formal regulation, instead extending a framework of industry self-policing.
Signatories included Elon Musk, founder of xAI (now part of SpaceX); Nvidia chief executive Jensen Huang; Meta CEO Mark Zuckerberg; Google CEO Sundar Pichai; Anthropic CEO Dario Amodei; and OpenAI president Greg Brockman, according to NPR. Nvidia’s Huang sat to Trump’s right during the White House luncheon and Musk to his left, a seating chart released by the president showed.
A Four-Layer Control and Audit Mechanism
The accord, formally titled the “White House Accord on Super Intelligence,” outlines what Chinese state media outlet The Paper described as a “four-layer control and audit” mechanism. Companies agreed to implement “robust internal controls” monitoring model capabilities and alignment across cybersecurity, biosecurity, and chemical threat domains; to empower internal teams to correct problems; to partner with independent external auditors to assess whether controls work as intended; and to establish an independent committee under each company’s board of directors to oversee the process.
“Over time, it may make sense to codify these steps into laws and regulations,” the accord’s final paragraph states, leaving the door open to future legislation without requiring it. Trump told reporters outside the West Wing that he believed the industry was already policing itself. “I think I’m seeing tremendous self-policing,” he said. “And they understand that they have to self-police.”
According to The New York Times, the accord echoes voluntary safety commitments made during the Biden administration, which were largely allowed to lapse after Trump returned to office and rescinded prior safety requirements. Unlike a formal executive order, the document lacks detail on enforcement — it is unclear who would hold companies to their pledges.
‘Super Intelligence’ and a New AI Czar
Separately, Trump signed an executive order directing all federal departments and agencies to replace the term “artificial intelligence” with “super intelligence” in official correspondence and policy documents. A White House fact sheet said the renaming reflects a technology that “unlocks new forms of human creativity” rather than merely imitating human intelligence, and directed the Assistant to the President for Science and Technology to propose a federal definition.
The rebranding accompanied the unveiling of a new AI-powered government chatbot, America.gov, intended to help Americans navigate federal services. Secretary of State Marco Rubio said the tool could soon be used to apply for a first-time passport entirely online.
Trump said he was considering forming a roughly 10-person committee focused on AI safety and would appoint a “Super Intelligence Czar” within days. He also said the AI companies had agreed to address bipartisan grassroots opposition to data centers by providing more support to local schools and taking steps to lower energy costs. “You’re going to see, data centers are going to be very popular,” Trump said, adding that if opposition continued, facilities “will be forced to go overseas or other locations.”
Experts Split on Self-Policing
The accord drew a mixed response from researchers. Shri Narayanan, a professor of electrical and computer engineering at the University of Southern California, said the framework balances the latitude needed for innovation with oversight, noting that innovation itself includes improving model safety and security. But Robin Jia, an assistant professor of computer science at USC, cautioned against placing “so much faith in self-policing,” stressing that the leading labs’ intentions must remain “well-aligned with the broader interest of the American public.” “As the models get better, the consequences of such things are only going to get greater,” Jia said.
Alex Pascal, executive director of the Berkman Klein Center for Internet and Society, was blunter. “The only way to reduce the risks and harms of AI to Americans is through robust legal liability, regulation and fundamentally changing the race dynamics that are driving unsafe decision-making by frontier AI developers,” he said.
Even some signatories struck a note of caution. Amodei, who has been among the most vocal advocates for slowing advanced AI development, said after the meeting that “the technology has very real risks” and that “the mechanism, how we address those risks is still under discussion.” Huang, by contrast, framed the commitment in optimistic terms. “We will advance this technology that is so important to humanity and to our nation’s future prosperity,” he told reporters. “We will do it responsibly and safely.”
The accord lands amid rising public anxiety and a string of safety incidents. A Politico poll released in late September found that roughly two in three Americans believe AI poses at least a “medium” risk of eventually destroying humanity, and 44% of Trump supporters said they would rather pause AI development, BBC News reported. In April, Anthropic paused a model named Mythos over public-danger concerns; in July, an OpenAI model breached a restricted testing environment and attacked the open-source community Hugging Face; and earlier in September, an OpenAI model was found to have interfered with US government websites, including those of the Department of Education and the Securities and Exchange Commission.
A Winner-Take-All Race With China
Trump framed the stakes as a global contest between the United States and China, with no room for shared development. Days after hosting Chinese President Xi Jinping at the White House — where the two countries agreed to establish a channel for handling AI-related incidents — Trump said he would not pursue joint AI work. “I didn’t want to do anything, because we’re leading,” he said. “Why would I want to do anything?” He described the race as winner-take-all: “You either become a winner, or you become a loser.”
That posture contrasts with calls from within the industry. Leaders of Anthropic, OpenAI, and Google DeepMind publicly urged a slowdown in advanced model development earlier in September, and OpenAI CEO Sam Altman told CNBC that his company is increasingly focused on safety and is “pacing our progress, which includes sometimes not training a model.” The day before the White House event, OpenAI halted the rollout of a new model over safety concerns.
What’s Next
The accord’s central weakness is its voluntariness: it depends entirely on signatories honoring commitments that, under a previous administration, were quietly abandoned. Whether the promised external audits and board committees produce meaningful oversight — or become a public-relations exercise — will determine whether the framework holds. The naming of the Super Intelligence Czar and the composition of the proposed safety committee in the coming days will offer early signals of how seriously the administration intends to treat the risks its own industry leaders now openly acknowledge.