We conducted cyber evaluations of Claude Mythos Preview and found that it is the first model to complete an AISI cyber range end-to-end. 🧵

Boris Cherny and colleagues conducted cybersecurity evaluations of Claude Mythos Preview, finding it to be the first AI model to successfully complete an AISI (AI Safety Institute) cyber range end-to-end. This represents a notable benchmark in AI cyber capability evaluations, as completing a full cyber range end-to-end demonstrates advanced autonomous offensive or defensive security task completion. The finding likely has significant implications for AI safety and capability thresholds tracked by the AISI.

So the concern over Mythos and cybersecurity seems warranted.
UK gov's Mythos AI tests help separate cybersecurity threat from hype
Mythos is very powerful, and should feel terrifying. I am proud of our approach to responsibly preview it with cyber defenders, rather than …
Cybersecurity analysis: Claude Mythos Preview had a 73% success rate on expert-level capture-the-flag challenges, which no model could finish before April 2025 (AI Security Institute)
Anthropic's Project Glasswing - restricting Claude Mythos to security researchers - sounds necessary to me

Source: Boris Cherny (X) | 2026-04-13

Related