Anthropic Mythos surpasses GPT-5.5 in cyberattack simulations by Q4 2027

OpenAI's GPT-5.5-Cyber completed a 32-step simulated cyberattack in 2 out of 10 test runs, while Anthropic's Mythos did so in 3 out of 10. Given Anthropic's rapid iteration pace and focus on AI security, Mythos is likely to improve further by Q4 2024.

Implied probability (Yes)

72%

Closes

2027-12-31