Mythos AI model surpasses GPT5.5 in exploiting Google Chrome vulnerabilities

Anthropic's Claude Mythos outperformed OpenAI's GPT5.5 in exploiting Google Chrome vulnerabilities. A new benchmark, ExploitBench, tested AI models' ability to exploit real-world vulnerabilities, with Mythos achieving significantly higher scores than GPT5.5. The benchmark measures staged exploitation outcomes, highlighting AI's growing capability in this area, though experts urge caution and emphasize the need for automated remediation.