Key points
- Six AI leaders signed a voluntary safety accord
- One line reads like our breach coverage
- The accord calls for outside checks but sets no deadline
President Donald Trump and six AI company leaders signed a voluntary safety accord at the White House on Tuesday. Asked whether it binds anyone, Trump called it "morally binding." That afternoon, he also backed faster data center expansion and directed federal agencies to call AI "super intelligence."
The signers were Alphabet's (GOOGL) Sundar Pichai, Meta's (META) Mark Zuckerberg, Anthropic's Dario Amodei, SpaceXAI's Elon Musk, Nvidia's (NVDA) Jensen Huang, and OpenAI President Greg Brockman.
After a month of covering AI agents accessing systems and sharing information without permission, I wanted to see what the companies had agreed to do. One line addressed exactly that behavior.
What did they actually sign?
The document is titled "White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities." It opens with a statement of responsibility: "every company is responsible for developing its own technology safely and in a way that builds trust with customers and the public."
It calls for four layers of oversight: controls on the models, an internal team to check those controls, an independent outside auditor or evaluator, and an independent board committee overseeing the work. The companies also say they'll "meet regularly to establish standards and best practices."
The final paragraph leaves legislation for later. "Over time, it may make sense to codify these steps into laws or regulations," it says. For now, the commitment is voluntary.
The name change comes through a separate executive order, which tells agencies to replace "Artificial Intelligence" and "AI" with "Super Intelligence" and "SI," and not to acknowledge the old terms. I'll keep calling it AI, since I'm not a federal agency.
Why does the hacking line matter?
The accord says companies should make sure their models "do not hack or access technical systems in unintended ways." That is already a concrete problem in the reporting we've covered.
Australia's prime minister said an OpenAI agent got into a Medicare portal, an incident the government learned about months later. We also reported on OpenAI agents posting ChatGPT users' images online. OpenAI and Anthropic have both reported agents acting outside their intended limits and hacking into other companies' systems.
Putting that behavior in the accord gives the companies a specific problem to address. What I'd want to see next is when the controls take effect and what evidence shows they work.
Who checks the work?
The accord calls for an independent outside auditor or evaluator, alongside internal controls and board oversight. What it doesn't specify is who those evaluators will be, when the checks must begin, or what happens if a company falls short. Trump said he's considering a 10-person board to police AI safety, but he didn't name anyone for it.
Bradley Tusk, CEO of Tusk Ventures, pointed out on CNBC that "almost all those same CEOs" said two weeks ago they should be regulated. What they want, he said, is "something that creates an equalizer for each other."
Zuckerberg's pitch was that it's a first step. "The idea isn't that this is the only thing that we'll ever do," he told reporters. "It is that this is a start and an accord that the whole industry can come to." Amodei, who called for pacing the frontier earlier this month, kept it short. "If we do this right, if we work with the president and everyone here we can win safely," he said.
Sen. Mark Warner wasn't buying it. "The president's response? To rename it and tell the companies developing it to regulate themselves," he said, and he's introduced a bill requiring testing before the most powerful models are deployed.
The other half of Tuesday was data centers, and that's where the politics get harder. Trump said the companies will make "massive contributions to communities" to win over towns fighting new builds. A Reuters/Ipsos poll from September 17 to 20 found 73% of people worry AI companies haven't done enough to prevent serious harm, and 55% said slowing AI down would be good. The midterms are November 3.
Who takes responsibility inside the companies?
For investors, the next useful detail is who takes responsibility for carrying out the accord. Company disclosures could identify the board committee overseeing AI safety, its members, and its authority to demand changes.
Anthropic's planned IPO gives investors a chance to press for those answers. Its prospectus is reportedly already drafted. I'd want the public version to explain who can act when an evaluator finds a problem, including whether they can delay a model's release. That would tell us more than another promise to develop AI safely.
