GPT-5.5 matches Claude Mythos in cyber attack tests, UK AI Security Institute finds

The UK AI Security Institute said OpenAI’s GPT-5.5 performed about on par with Anthropic’s Claude Mythos Preview in advanced cyber evaluations.

In some runs, GPT-5.5 completed a multi-stage network attack simulation, showing stronger end-to-end cyber capability.

Researchers also found it could be hit by a universal jailbreak that enabled malicious cyber requests, exposing a safety gap.

The findings add to concerns that frontier AI models are becoming more capable of real-world cyberattacks while remaining bypassable.