Anthropic changes its safety promise amid Pentagon dispute

Anthropic announced it is replacing its self‑imposed Responsible Scaling Policy with a more flexible, nonbinding Frontier Safety Roadmap. The move comes as the company faces Pentagon pressure and a reported ultimatum tied to a potential $200 million contract.

Anthropic announced a shift from its earlier Responsible Scaling Policy to a more flexible, nonbinding Frontier Safety Roadmap. The company said its prior pause-oriented commitments could hinder competitiveness and that the new framework will set public goals rather than fixed promises. The change was disclosed the same week Anthropic met with U.S. Defence Secretary Pete Hegseth, who reportedly pressed the company over its safeguards. Anthropic also signaled it will keep certain positions, notably on AI‑operated weapons and mass domestic surveillance. Key points: - Anthropic replaced its prior policy that called for pausing model training if control could not be ensured with a nonbinding Frontier Safety Roadmap of public goals. - Defence Secretary Pete Hegseth reportedly gave Anthropic an ultimatum to roll back some safeguards or risk losing a reported $200 million Pentagon contract and face other government measures. - Anthropic stated it will not abandon concerns about AI‑controlled weapons and mass domestic surveillance. - A reported Friday deadline was given during the meeting; further government actions, including supply‑chain designations, have been mentioned by officials. Summary: Anthropic has made its safety commitments more flexible while under reported pressure from Pentagon officials about access to the company’s technology. The immediate procedural step reported is a near‑term deadline set by the Defence Secretary; the broader consequences and next moves are undetermined at this time.