Dear Colleagues,

In July, autonomous agents running on an unreleased frontier model broke into Hugging Face’s production systems and tried to erase their tracks. A survey of 272 experts puts the chance of AI causing over a million deaths or $100 billion in damages before 2030 at roughly one in five on the current course. Even a far smaller failure would hit every developer. Three Mile Island injured no one, yet the U.S. authorized no new reactor construction for more than three decades. A similar stall in AI would reach the chipmakers, utilities, data-center debt, and retirement savings that institutional investors hold across the industry, and no portfolio can diversify that away. Even so, several developers have recently made their safety commitments contingent on what competitors do, leaving everyone hostage to the least careful one. Our letter asks them to restore unconditional safety thresholds, back a binding international body to enforce shared standards, and disclose safety evaluations and incidents to investors. We invite your institution to sign.

The intended recipients are the boards and CEOs of global frontier AI developers, including the boards of their foundations where applicable. Formative will send the letter on behalf of all signatories, and coordinate resulting engagemets. The deadline to sign on is November 30.