OpenAI is preparing to launch GPT-5.5-Cyber. It is to be a specialized cybersecurity model restricted to vetted cyber defenders. CEO Sam Altman announced the limited rollout will happen within days, though technical details remain undisclosed. It not only follows a similar deployment to GPT-5.4-Cyber, but obviously follows in the footsteps of Anthropic’s Claude Mythos as well. Interest in these types of LLMs, perceived to be too dangerous to release outright, has significantly increased.
The specialized cybersecurity model will not be available to the general public, CEO Sam Altman said. The rollout will instead target a select group of trusted “cyber defenders,” with access expected “in the next few days,” Altman stated on X. No technical details or specifications have been released; the name suggests what’s simply a specialized version of GPT-5.5, OpenAI’s recently released flagship model focused on agentic capabilities.
“We will work with the entire ecosystem and the government to figure out trusted access for Cyber,” Altman added. Who exactly will receive early access remains unclear. Previous trusted access programs from OpenAI have involved vetted professionals and institutions, a model it already applied with the earlier GPT-5.4-Cyber, which was rolled out to thousands of security specialists through its Trusted Access for Cyber program.
Part of a wider trend in controlled AI releases
The staggered rollout is thus nothing new. OpenAI has released previous focused models selectively, and this month introduced GPT-Rosalind, a purpose-built life sciences model for biology research and drug discovery, under a similar restricted approach. Anthropic followed the same playbook with Claude Mythos and caused a significant stir. It’s pseudo-launch was accompanied by considerably more fanfare, and a data leak that accidentally exposed draft documents fanned the flames of fear. Microsoft has since confirmed it is integrating Mythos into its Security Development Lifecycle.
The White House has also taken an active interest in the Mythos rollout and has reportedly opposed plans to expand access further. The U.S. government is citing cybersecurity concerns and worries that increased demand could limit Washington’s own ability to use the system, according to The Wall Street Journal. OpenAI’s explicit mention of working with government on “trusted access for Cyber” puts GPT-5.5-Cyber squarely in that conversation. OpenAI had already flagged the growing cyber risk profile of frontier models earlier, establishing its Frontier Risk Council to manage them.