Voices
UK AISI Publishes Claude Mythos Preview Cyber Capabilities Evaluation
The UK AI Safety Institute published its evaluation of Claude Mythos Preview's cybersecurity capabilities, assessing the model Anthropic has restricted to ~50 organizations via Project Glasswing. Mythos can autonomously discover and exploit zero-day vulnerabilities 'across every major operating system and browser,' according to Anthropic's own characterization. The AISI evaluation provides the first independent government assessment of frontier model offensive cyber capabilities — a template likely to be replicated by other national AI safety bodies.
Source
↳ Follow the thread