New benchmark shows Claude Mythos and GPT-5.5 can develop real browser exploits autonomously

Researchers at Carnegie Mellon University have developed a new benchmark to assess AI agents' ability to exploit real vulnerabilities in Google's V8 engine. The benchmark reveals that Mythos significantly outperforms GPT-5.5, albeit at a cost twelve times higher. This benchmark is crucial for evaluating AI's capabilities in creating real browser exploits, providing precise data on performance and cost. It highlights the substantial gap between different models in this domain.