The four rules in the White House AI commitment, and the missing fifth: penalties

The Verge's Jess Weatherbed reports the full text of the AI safety commitment announced by President Trump, as posted by White House adviser David Sacks. It was signed by Sundar Pichai of Google, Dario Amodei of Anthropic, Mark Zuckerberg of Meta, Greg Brockman of OpenAI, Elon Musk of xAI and Jensen Huang of Nvidia. It is framed as a "morally binding" Joint Commitment on Frontier Responsibilities.

The commitment sets four rules. First, robust internal controls during training and deployment, covering cyber, biological and chemical risks and unintended system access. Second, an empowered internal safety team. Third, an independent external auditor or evaluator. Fourth, an independent board committee overseeing those teams. The document says these steps "may" become law over time. It contains no penalties for breach.

The inclusion of unintended system access in the first rule is notable given a run of incidents in which agents reached systems they were not supposed to, including OpenAI's Hugging Face episode. The external evaluator echoes industry discussion of embedding outside safety evaluators, and the board committee responds to questions raised in debates such as whether AI labs can police themselves.

Executives had earlier discussed how to pace the frontier, and the commitment can be read as a minimal shared baseline. Critics, including WIRED in its pinky-swear critique and Brian Barrett's seat-belt essay, argue that without enforcement it falls short of regulation.

Several details would show whether the four rules have practical force: naming the external evaluators and the scope of their assessments, publishing board committee reports after incidents and adding penalty language if the commitment is revised. The phrase "may" become law leaves open whether this is a roadmap toward legislation or a substitute for it.

Even without enforcement, safety and security teams inside the labs can use the four rules as an internal checklist, and outside observers can measure each company against them. Reactions from the executives are covered in what AI leaders are saying. More in our regulation coverage and AI coverage.