AI exploits

New benchmark testing shows Claude Mythos and GPT-5.5 can autonomously develop real browser exploits without human assistance. This capability represents a significant escalation in AI-driven cybersecurity threats that could impact enterprise security budgets and defensive technology investments.