Epoch AI evaluates Claude Mythos, leading in vulnerability exploitation but overestimating discovery advantage

Epoch AI released a report evaluating the Claude Mythos series models, stating that Mythos Preview leads the linear trend in vulnerability exploitation development by about 7 months and outperforms OpenAI GPT-5.5 by about 3 months. The report believes that Mythos' advantage in vulnerability discovery is overestimated, as its performance in curl testing did not exceed traditional tools, and the surge in CVE reports from April to May was mainly due to Anthropic's investment of $100 million in API quotas to Glasswing. The practical advantage of Mythos lies in its low false alarm rate and precise hazard classification.