Anthropic Mythos 5.2 Hits 76.3% on Cyber Test
Anthropic has unveiled Claude Mythos 5.2, a cutting-edge AI model for cybersecurity engineering, which marks a significant advancement with its 76.3% score on the CyScenarioBench. This latest iteration outperforms its predecessor, Mythos 5.1, by 14.6 percentage points. It aids security teams in locating and fixing software vulnerabilities while maintaining restricted access due to potential misuse by malicious actors. The model also excelled in Anthropic’s internal CVE Rediscovery evaluation, scoring 89.6%, surpassing the previous version and Opus 5.5.
Notably, Claude Mythos 5.2’s accomplishments are more relevant when compared internally than with other vendors due to differing test tasks. The release is a continuation of Anthropic’s focus on improving vulnerability detection and repair processes. Enterprise users can leverage Claude Security for extensive integration without full model access, maintaining human oversight.
For an in-depth look at how Anthropic Mythos excels in cyber tests, visit the full article below.