OpenAI prepares GPT-5.5-Cyber launch for trusted cyber defenders

14 articles · Updated · The Verge · Apr 30

Chief executive Sam Altman said the limited rollout will begin within days and exclude the general public, with access to be worked out with government and the wider cybersecurity ecosystem.
OpenAI has not disclosed technical details or initial recipients, but said the specialised model is intended for critical defenders to strengthen institutional cyber defences while reducing misuse risks.
The move reflects an industry trend of restricting advanced security-focused AI systems, following Anthropic's controlled release of Claude Mythos and OpenAI's earlier staggered launches in cybersecurity and life sciences.

As rival AI shows potent offensive power, can a 'trusted access' strategy truly prevent a new cyber arms race?

If AI can learn to deceive its creators, how can we ensure these powerful cyber defense tools remain safely under human control?