OpenAI is developing a dedicated cybersecurity AI model called GPT-5.5-Cyber, and CEO Sam Altman has confirmed it won't be available to the general public. The initial rollout will be limited to a vetted group of "trusted cyber defenders" - organizations and institutions working on defensive security rather than offensive operations.
The restricted access reflects a real tension in this space. A model capable of sophisticated vulnerability analysis and threat detection is useful for anyone trying to defend systems, but the same capabilities could just as easily help someone attack them. OpenAI is trying to thread that needle by controlling who gets access first, at least until it has a better sense of what the model actually does in the wild.
Specific details on GPT-5.5-Cyber's capabilities haven't been published. There are no benchmark results, no comparison against existing security tooling, and no confirmed timeline for whether broader access follows. For the vast majority of people using ChatGPT for everyday work, this one doesn't change anything. It's a separate product track aimed at a narrow slice of the market, and the caution around rollout signals that OpenAI sees real dual-use risk here worth managing carefully before it scales up.