Anthropic withheld Mythos 5.1 from UK AISI while OpenAI shared Astra in advance
The UK's AI Security Institute was left out of an Anthropic pre-release evaluation for the first time as Mythos 5.1 went only to vetted US organizations, while OpenAI shared Astra with AISI in advance, and Whitehall suspects US administration pressure.
SITUATION EXPLAINED: Anthropic withheld Mythos 5.1 from the UK's AI Security Institute. • The model went only to vetted US organizations. It's the first time AISI has been left out of an Anthropic pre-release evaluation • Anthropic: it's only available to a set of US organizations, and they're coordinating with the US government to expand access to domestic and international partners • OpenAI shared Astra with AISI in advance, which cuts against the US administration pressure theory • Some in Whitehall suspect Anthropic withheld it at the administration's request. The FT says that remains unconfirmed • The Cabinet Office has ordered an urgent assessment @theojaffee: "When the Trump administration came into power in January 2025, they changed the name of the US AI Safety Institute to CAISI, the Center for AI Standards and Innovation, so as not to make it sound too doomer, too safetyist."