Watchtower

Anthropic replaced the original December 2025 Frontier Compliance Framework (FCF) with a new version on its trust center. The update was not publicly announced or noted in the trust center's update log. The new document includes a changelog at the end, though it offers limited specifics. The previous version of the FCF seems to no longer be published anywhere.

California's SB 53 requires frontier AI companies to "clearly and conspicuously publish” material modifications to their safety frameworks within 30 days. The update was made available through the trust center but was not accompanied by a separate disclosure.

The revised framework modified the risk thresholds for all risk categories. Among the changes were the addition of evaluation thresholds for harmful manipulation capabilities and a narrowing of the scope of AI R&D from general autonomous capabilities to a smaller set of specific fields.