GPT-5.5 Matches Mythos Preview in Cybersecurity Benchmarks

GPT-5.5 has pulled even with Anthropic's Claude Mythos Preview in cybersecurity capabilities, according to a new analysis from the AI Security Institute. The model is now the second to successfully solve a multi-step cyberattack simulation.

The AI Security Institute previously evaluated Mythos Preview back in April and found it represented a notable jump in cyber performance compared to earlier models. GPT-5.5 now reaches a similar level.

Two AI models can now navigate complex, multi-step cyberattack scenarios — a benchmark that until recently no model could clear. The fact that a second model has caught up quickly signals that frontier cyber capabilities are spreading across the industry, not staying confined to a single lab.

The findings raise ongoing questions about how rapidly AI-powered offensive cyber capabilities are advancing and what guardrails need to keep pace.