AI accord: Trump says tech firms agree to self-policing

US President Donald Trump says the leaders of the largest artificial intelligence companies have signed a voluntary accord that commits them to internal and external reviews of their AI development.

The announcement came on Tuesday after a meeting at the White House that was meant to address Americans' fears about the technology. Trump and House Speaker Mike Johnson met with executives from firms that are spending hundreds of billions of dollars on AI infrastructure. That build-out includes data centers in communities across the country, and those projects have drawn grassroots opposition from both parties.

"I think I'm seeing tremendous self-policing. And they understand that they have to self-police," Trump told reporters at an impromptu press conference outside the West Wing.

Who signed and what they agreed to

Besides Trump, the accord was signed by Anthropic CEO Dario Amodei, Google CEO Sundar Pichai, Meta CEO Mark Zuckerberg, OpenAI President Greg Brockman, Nvidia CEO Jensen Huang and Elon Musk, founder of xAI, which is now part of SpaceX.

Trump posted the text on his social media platform, Truth Social, late on Tuesday. It sets out four voluntary steps. The companies will put "robust internal controls" in place and work with an "independent external auditor" to check whether those controls work. Each company will also set up a committee within its board of directors to review the reports from internal and external auditors.

The document leaves room for regulation later. "Over time, it may make sense to codify these steps into laws and regulations," it says.

Trump said about 10 people would sit on a committee to "watch over the whole enterprise" and called the accord "morally binding." He added that he would name someone to oversee the agreement in the coming days, after consulting with industry. Some of the measures in the accord are things the companies already do in some form, or have promised before.

Experts are divided

Amodei, one of the most vocal advocates for slowing AI development, struck a cautious note. "The technology has very real risks," he said. "And, you know, the mechanism, how we address those risks is still under discussion."

Shri Narayanan, a professor of electrical and computer engineering at the University of Southern California, said the accord tries to balance room for innovation with regulation. In his view, that innovation covers not only model capabilities but also work on model safety and security.

Robin Jia, an assistant professor of computer science at the same university, is less convinced by "so much faith in self-policing." He said the labs' intentions must be "well-aligned with the broader interest of the American public" and pointed to recent disclosures of hacking incidents. "As the models get better, the consequences of such things are only going to get greater," he said.

Alex Pascal, executive director of the Berkman Klein Center for Internet and Society, criticized the vague language. He argued that the only way to reduce AI harms is "through robust legal liability, regulation and fundamentally changing the race dynamics" that drive unsafe decisions by frontier AI developers.

Chatbot, data centers and China

Earlier the same day, Trump unveiled America.gov, an AI chatbot meant to help Americans navigate government services. He promised not to back "guardrails" or other limits. "I will never stifle the growth of a technology that will be bigger than the industrial revolution," he said. Secretary of State Marco Rubio said the chatbot could soon let people apply for a first-time passport fully online.

Trump also said the executives had agreed to ease local resistance to data centers, for example by giving more money to local schools or cutting energy costs. If opposition continues, he suggested, the projects "will be forced to go overseas or other locations."

OpenAI CEO Sam Altman told CNBC that the company is increasingly focused on safety, a day after OpenAI halted the rollout of a new model over safety concerns. "We are pacing our progress, which includes sometimes not training a model," he said.

Trump framed AI as a race with China with no "second place." The two countries agreed last week to set up a channel for AI-related incidents, but Trump said he did not want to cooperate too much: "Why would I want to do anything?"

Our Take

For security teams, the key word here is "voluntary." The accord describes controls, audits and board oversight, but it does not define what those controls must cover or what happens if a company falls short. Calling it "morally binding" suggests there is no legal enforcement behind it for now.

The timing matters. Regulators are already looking at the same companies, with the FTC investigating OpenAI and Anthropic over consumer risks, and the debate over liability for AI agents is heating up. It is worth watching who Trump picks to oversee the accord, whether the external auditors publish anything, and whether the accord's own hint about codifying these steps into law ever turns into actual legislation.