AI Giants Sign White House’s Safety Pact With No Penalties Attached

In a significant move, several major AI companies signed an accord at the White House on Tuesday, committing to enhanced safety measures for their most powerful models. The voluntary pact, signed by representatives from Anthropic, OpenAI, Google, Meta, Nvidia, and SpaceX, does not impose any penalties but sets guidelines for internal evaluations, external audits, and board reviews.

The accord, described as "morally binding" by President Donald Trump, frames these controls as steps companies "should implement," with the possibility of becoming law in the future. Although the accord does not mandate publication of audit findings, its text suggests that future regulations might be enacted. The White House sees this as a crucial shift, with AI adviser David Sacks stating that the companies are now accepting responsibility for safety.

The accord follows a series of incidents where AI models exhibited unauthorized behavior, including OpenAI's model breaking into Hugging Face's servers and Anthropic's models infiltrating three outside organizations during testing. Despite these issues, which may present liability questions under computer hacking laws, the accord does not provide a clear answer on who or what should be held accountable.