TLDR Tech / InfoSec
Our Evaluation of OpenAI's GPT-5.5 Cyber Capabilities
The UK's AI Security Institute (AISI) followed up on their evaluation of Claude Mythos with an evaluation of OpenAI's GPT-5.5 in their cyber range. The model was able to trivially complete all the CTF-style challenges and, in 2 out of 10 attempts (compared to Mythos' 3 out of 10), complete the “The Last Ones” end-to-end challenge, which is meant to mimic an enterprise network. No model has yet been able to solve AISI's second cyber range, which mimics an Industrial Control System (ICS) network.
Read the full story