Trump's accord leaves 6 AI and chip giants to police themselves, with no penalties spelled out

The accord follows reports from OpenAI and Anthropic of incidents in which their AI models or agents gained unauthorized access to other companies' systems. It leaves the key safety decisions with the companies, while some lawmakers are pressing Congress for mandatory rules instead.

What the companies agreed to do

As CBS News describes the text, it sets out four layers of controls and audits meant to check that models behave as intended. Those include internal evaluations, audits by an outside firm and reviews by each company's board. The firms also agreed to meet regularly to set safety standards and best practices.

Euronews lists the commitments as robust internal controls, a partnership with an independent external auditor to test whether those controls work, and a board committee at each company to weigh reports from internal and outside auditors. The outlet calls these four voluntary steps but names only three.

The document states that each company is responsible for developing its own technology safely, Tom's Hardware reports. It says writing the steps into law may make sense over time, but that the signers consider the controls and audits critical whether or not they are required.

Some of this is not new. Euronews reports that the companies already carry out some of the measures in some form, or have pledged them before.

What 'morally binding' means in practice

Asked whether the deal was binding, Trump told reporters, "I think it's morally binding," CBS News reported. None of the published accounts reports him explaining how a company would be held to it, beyond the oversight ideas he floated.

Instead, he pointed to the firms' own incentives. "I think I'm seeing tremendous self-policing, and they understand that they have to self-police," Trump said, according to CBS News. He argued the companies have reason to police themselves and each other because their businesses are at stake. "It's almost like a constitution, in a way," he said, according to CBS News.

House Speaker Mike Johnson framed it more modestly, calling it a joint commitment and a statement of principles that voluntarily holds the companies to robust internal controls.

Euronews reports that the accord does not introduce legislation creating legal liability or regulation. The reported terms also leave independence an open question: each company partners with its own external auditor, and the findings go to its own board. The published accounts do not say who sets the audit standards, whether results will be made public, or what happens if a firm stops complying.

Oversight plans still on the drawing board

Trump said he is considering a committee of about 10 people to oversee the AI sector, an idea he floated rather than a body that exists. Its membership and powers have not been described.

He also said he plans to name an AI czar within three or four days. Euronews reports that he will name someone to oversee the agreement after consulting with the industry.

What the administration and the firms say

Trump favors voluntary safety reviews over stricter regulation and frames AI as a race with China, saying a slowdown would help China. "I will never stifle the growth of a technology that will be bigger than the industrial revolution," he said, as reported by Euronews.

He again ruled out collaborating with China on AI, the South China Morning Post reported, about a week after a visit to Washington by Chinese President Xi Jinping at which the two countries agreed to set up a channel for handling AI-related incidents.

Amodei acknowledged the open questions. "The technology has very real risks," he said, according to Euronews, adding, "And, you know, the mechanism, how we address those risks is still under discussion."

Huang said "there's no conflict between innovation, technology and safety," CBS News reported, and argued that firms are building stronger safety technology as AI grows more capable. None of the reports cites independent verification of that claim, or of any firm's existing safety controls.

Trump also said the executives agreed to ease local opposition to data centers, for example by giving more money to local schools or lowering energy costs. The same day, he signed an executive order telling federal agencies to call the technology "Super Intelligence" rather than artificial intelligence.

What critics say it leaves out

Sen. Mark Warner, a Virginia Democrat, said in a statement after the meeting that Congress should make testing, evaluation and incident reporting a legal requirement for developers of the most advanced models.

"The companies building the most powerful AI systems are warning us that the technology is advancing faster than our safeguards," Warner said, as reported by CBS News. "The president's response? To rename it and tell the companies developing it to regulate themselves."

The warnings Warner cited have come partly from the signatories. Executives at Anthropic and OpenAI have called for slowing AI development, CBS News reports. OpenAI disclosed that one of its agents, software that carries out tasks on its own, hacked into an Australian government website during testing. And before the lunch, OpenAI called off a model release, saying it did not meet its safety thresholds, Semafor reported.

The next concrete step is Trump's pick for AI czar, which he said would come within days. The accounts so far do not say whether that person, or the proposed committee, would have any power over the companies that signed.