The new specialized AI model is being rolled out to a select group of vetted professionals to enhance defensive cybersecurity workflows.
OpenAI is rolling out GPT-5.5-Cyber, a specialized AI model designed for vetted security professionals to conduct advanced defensive cybersecurity workflows, initially to a select group of "critical cyber defenders."
OpenAI has begun the limited rollout of its new frontier cybersecurity model, GPT-5.5-Cyber, to a handpicked circle of "critical cyber defenders." CEO Sam Altman announced on X that the model will not be available to the general public, with initial access restricted to trusted entities. This strategic deployment aims to rapidly help secure companies and infrastructure by allowing institutions to shore up their cyberdefenses.
The rollout is set to commence "in the next few days," with OpenAI planning to collaborate with the broader ecosystem and government to establish trusted access protocols for the specialized AI. GPT-5.5-Cyber is engineered to identify flaws before they can be exploited, offering capabilities in penetration testing, bug discovery, exploitation, and malware analysis.
The introduction of GPT-5.5-Cyber comes with OpenAI's most robust set of safeguards to date, specifically designed to mitigate misuse while ensuring access for beneficial work. These measures include authenticated access controls, comprehensive monitoring systems, and stricter limits on cyber workflows deemed higher risk. External experts have rigorously tested these safeguards to assess their robustness.
This controlled release strategy mirrors Anthropic's earlier rollout of its Claude Mythos model to a limited number of organizations. OpenAI's approach emphasizes responsible deployment, building on cyber-specific safeguards first introduced with GPT-5.2 and continuously refined in subsequent versions. The company is working closely with partners and customers to address safety and security requirements for large-scale deployment, particularly for API access.
While GPT-5.5-Cyber targets a specialized audience, OpenAI also announced the broader rollout of GPT-5.5 to its paid subscribers, including Plus, Pro, Business, and Enterprise users, across ChatGPT and its coding assistant Codex. This latest AI model significantly enhances capabilities in coding, computer usage, and deeper research. OpenAI President Greg Brockman highlighted the model's ability to tackle unclear problems with less guidance, setting a new foundation for computer-based workflows.
The UK's AI Security Institute has already lauded GPT-5.5-Cyber as "one of the strongest models we have tested on our cyber tasks," noting it is only the second system to complete one of its multi-step attack simulations end-to-end. This underscores the model's advanced capabilities in both breaking and fixing systems, emphasizing the critical importance of its controlled distribution. OpenAI views GPT-5.5 as an incremental yet vital step towards AI that can address some of the world's most challenging problems, particularly in cybersecurity.
