Agents
104d agoAI Agents Autonomously Develop Browser Exploits
Researchers at Carnegie Mellon University built a benchmark testing AI agents against Google's V8 engine vulnerabilities. Claude Mythos outperformed GPT-5.5 in developing real exploits, though it cost twelve times more to run. This result proves LLMs can now automate complex security breaches.
The Signal
Practitioners must prioritize hardened browser environments to mitigate these autonomous threats.