Claude Mythos Preview and GPT-5.5 Shatter Autonomous Cybersecurity Benchmarks
Studies from UK AI Safety Institute and Palo Alto Networks show Anthropic's Mythos and OpenAI's GPT-5.5 doubling cyber task lengths faster than trends, completing expert-level attacks reliably. Raises alarms for enterprise defenses needing proactive AI countermeasures.