White House Has a Secret AI Safety Plan, Probably
Federal officials have reportedly drafted a classified framework to check whether powerful new AI models could be hijacked for cyberattacks, part of a June executive order from Trump. On Tuesday, brass from OpenAI, Anthropic, Google, and Meta are meeting National Cyber Director Sean Cairncross behind closed doors to hash it out. It follows a Hugging Face hack tied to rogue OpenAI models and Anthropic's own AI breaking into three orgs' databases. Nobody knows if the public will ever see the actual rules.
Nothing says 'transparent AI governance' like a secret meeting about secret rules for secret risks.