OpenAI Rolls Out Less Restricted GPT-5.5-Cyber to Vetted Security Defenders
Similar Articles
New AI Models Show Advanced Cybersecurity Capabilities in UK Safety Tests
OpenAI Releases GPT-5.5 Model, Focused on Efficiency and Enterprise Use
AI Firms Brief Congress on Advanced Cybersecurity Models and Risks
U.S. Cyber Command to Deploy Top AI Models for Cyber Operations
CISA Lacks Access to Anthropic's Security AI Model as Other Agencies Use It for Testing
OpenAI has begun a limited preview of a more permissible version of its GPT-5.5 model, called GPT-5.5-Cyber, for vetted cyber defenders. The model is designed to assist with tasks like hunting for bugs and reverse engineering attacks while blocking malicious uses. Its capabilities appear to be competitive with similar models like Anthropic's Mythos, which has also been made available to select organizations.
Facts First
- OpenAI is previewing GPT-5.5-Cyber for vetted defenders responsible for critical infrastructure security.
- The high-tier version has fewer guardrails than the public model, allowing bug hunting and malware analysis but blocking credential theft or malware writing.
- The model's abilities are reportedly on par with Anthropic's Mythos, which recently completed a complex simulated attack in 3 out of 10 test runs.
- Anthropic has granted Mythos access to approximately 40 organizations, some of which also participate in a model-testing information-sharing group.
- The White House is discussing executive actions regarding federal involvement in future AI model rollouts.
What Happened
OpenAI announced it is rolling out a more permissible version of GPT-5.5, known as 'Spud' or GPT-5.5-Cyber, to vetted cyber defenders. The limited preview is available to approved members of OpenAI's Trusted Access for Cyber program, specifically those responsible for securing critical infrastructure. These defenders receive a version with fewer guardrails, allowing tasks like hunting for bugs, studying malware, and reverse engineering attacks, while blocking tasks such as credential theft and writing malware. Another version of GPT-5.5 is available to other program members for tasks like understanding unfamiliar code or reviewing patches.
Why this Matters to You
If you rely on services from banks, utilities, or hospitals, this development may lead to stronger defenses for the critical infrastructure that underpins your daily life. The tools could help security teams find and fix vulnerabilities faster, potentially reducing the risk of disruptive data breaches or service outages. The White House's discussions on federal involvement in future model rollouts suggest this area of AI development is likely to receive increased government attention.
What's Next
The performance of GPT-5.5-Cyber and similar models like Anthropic's Mythos will continue to be tested and compared. The U.K. AI Security Institute's recent report, which showed both models completing a complex simulated attack, indicates this benchmarking is already underway. The White House's executive action discussions could shape how these powerful AI tools are deployed and regulated in the future.