President Trump and the heads of six AI companies signed a voluntary safety accord at the White House on Tuesday, and five of those six companies had already made a voluntary AI safety pledge to a president before.
Here is what the accord actually establishes:
- The signers, according to Al Jazeera, were Anthropic’s Dario Amodei, OpenAI’s Greg Brockman, Google’s Sundar Pichai, Meta’s Mark Zuckerberg, xAI’s Elon Musk and Nvidia’s Jensen Huang.
- They pledged “robust” controls to monitor their models, an internal team to make sure those controls work and that “any issues are remediated,” and a partnership with independent outside auditors.
- Euronews adds a board-level committee at each company to review what the auditors find.
- There is no penalty for breaking any of it. The text says only that codifying the measures into law “may make sense.”
Trump called it “morally binding.” That is a phrase, not a mechanism.
Now the part the coverage skips. On July 21, 2023, the Biden White House announced that seven companies had agreed to “internal and external security testing” of their AI systems before release, plus information sharing, watermarking of AI-generated content and public reporting on capabilities and limits. Anthropic, Google, Meta and OpenAI were on that list. FedScoop’s account has Nvidia joining in a later round. Only xAI is a first-time signer this week.
So five of the six have now made two voluntary promises to two presidents of opposite parties. External testing is not new here either; the 2023 text already promised it. What the 2026 accord adds is the independent auditor, the board committee and a sentence conceding that law might someday be sensible.
That 2023 fact sheet ended with a promise of its own: the administration would “pursue bipartisan legislation.” Three years on, the new accord still describes legislation as something that “may make sense.” Whatever else you conclude, the earlier pledge did not lead to the law it said it was a first step toward.
Did the earlier pledge do anything? FedScoop reported that as of November 2024 only some of the signers, among them OpenAI, Nvidia, Amazon and IBM, publicly said they would keep honoring it, while Anthropic and Microsoft declined to comment. MIT Technology Review, in the same account, found some progress on red-teaming and cybersecurity and less on public reporting. That is a mixed grade, and nobody had the authority to give a failing one. Microsoft, an original 2023 signer, was absent from the meeting this time, and nothing in the accord says what happens to a company that quietly drops out.
Trump’s case for skipping enforcement was that “we automatically have regulation with the Department of Justice, the FBI, all of that,” as Al Jazeera quoted him. This is the weakest link in the argument. The Justice Department and the FBI act after something has gone wrong, and only when a statute has been broken. A pre-release safety regime is a different thing: someone with the power to say no before the model ships. Euronews noted that an OpenAI agent hacked an Australian government website during testing. That is the category of event a voluntary control is supposed to catch, and the accord does not say who checks whether it did.
UNSW’s Toby Walsh put the objection in one line to Al Jazeera: “What other trillion-dollar industry marks its own homework?” We think that is the right question, and the honest answer is none we would trust. The accord does not say how the auditors are chosen, who sets their standards or whether anyone outside the companies sees what they find, and a committee each company designates for itself is homework with an extra signature on it.
Our read: a voluntary pledge is fine as a floor and useless as a ceiling, and the White House is selling it as the ceiling. Three things would change that. The auditors have to be named. Their reports have to be public, in whole or in part. And a company that walks away has to face something more than a bad news cycle. Trump also floated a committee of “maybe 10 people” to watch over the whole enterprise, per ABC News, and did not say who they would be.
Watch for the auditor list first. If it does not appear within a few months, the 2026 accord will have followed the 2023 one, and the next president will get to sign a third.