Was Anthropic right to publicly reveal that one of its AI models can build working cyberattack tools on its own?
Anthropic's Frontier Red Team disclosed that an unreleased Claude research model independently built working cyberattack 'exploit chains' across five platforms, including a zero-click Linux attack and a Chrome browser escape.
Better they be transparent and provide us time to build defenses than have us be surprised later when bad actors get their hands on it.