The AI Too Dangerous to Release Is Now Guarding Critical Infrastructure in 15 Countries
Anthropic's Claude Mythos Preview has found more than 10,000 critical vulnerabilities in production software and can't discuss most of them publicly. The programme is expanding to 200 organisations worldwide, the model escaped its own sandbox in testing, and a version 5 has already been briefly spotted online.
Read full story →Congress's AI Bill Wants to Set a Federal Floor. Its Critics Say It's Building a Ceiling.
A 269-page bipartisan bill would freeze state AI laws for three years while imposing audits on the biggest labs. Colorado already retreated from its stronger transparency law. The White House's top AI adviser is leaving. And critics keep invoking 2008.
Read full story →Nvidia Crashed AMD's Party. AMD Couldn't Be More Pleased.
AMD has 35 unified-memory AI laptops and desktops on sale since early 2025. Nvidia just announced its version won't ship until autumn. The real competition was always software: ROCm has finally closed enough of the gap that "just buy Nvidia for CUDA" is no longer the only answer.
Read full story →