Critics: Anthropic's compliance framework has no provisions for internal model risks
DKokotajlo · x · 2026-10-11
Safety policy researcher Nathan Calvin points out that Anthropic's frontier compliance framework contains no provisions whatsoever for managing risks from its internal models. Anthropic says it expects its approach to evolve and the framework to be updated, but months after its frontier release, nothing has changed.
Garrison Lovely adds that since January, California law has allowed AI companies to bind themselves to safety plans they are legally compelled to follow—yet even the "safety" company Anthropic publishes two versions of its safety plan, with the weaker one being the legally binding one.
Related event: Critics Say Anthropic's Safety Commitments Have Legal Loopholes(3 posts)→
More from Companies & People
- a16z speedrun hosts first SpeedHacks hackathon with expedited investor interviews — emax · 2026-10-11
- Breaking Down Higgsfield's GTM and Launch Strategy — VibeMarketer_ · 2026-10-11
- The Zvi: OpenAI employees' silence on chilling-effect events is 'the dog not barking' — birchlse · 2026-10-11
- pmitu: distribution is now the hardest feature to build — alexmacgregor__ · 2026-10-11
- High schooler with 3 NeurIPS papers applies for AI research internship, remote only — DominiqueCAPaul · 2026-10-11
- Musk courts chip engineers as Terafab targets 10 chip designs a year — elonmusk · 2026-10-11