OpenAI launches GPT-5.5-Cyber: a new era of cybersecurity amid Anthropic restrictions
On June 22, OpenAI officially released the full version of its specialized model, GPT-5.5-Cyber, designed for detecting, verifying, and fixing vulnerabilities. This move is a strategic response to recent restrictions imposed around Anthropic and marks a new stage in the development of AI tools for cybersecurity. The model is part of the Daybreak program and will not be available to the general public — it is created exclusively for verified security professionals who require advanced capabilities for defensive work.
A Window of Opportunity for OpenAI
The market situation changed after June 9, when Anthropic granted access to two models from the Claude family — Fable 5 and Mythos 5. However, on June 12, the developer was forced to disable them due to a directive from the U.S. government under export control regulations. This sparked a wave of discontent: on June 23, the company Legion filed a lawsuit against the U.S. government, claiming that losing access to Anthropic's models disrupted their legal tools. OpenAI, unlike its competitor, chose a more cautious path: the company pre-arranged checks with federal agencies and restricted access to GPT-5.5-Cyber to verified users only.
Technical Specifications and Performance
According to OpenAI, GPT-5.5-Cyber demonstrates impressive results on specialized benchmarks. On CyberGym, the model achieved 85.6% compared to 81.8% for the standard GPT-5.5. On ExploitGym, results were 39.5% versus 25.95%, and on SEC-bench Pro, 69.8% versus 63.1%. These figures reflect the model's ability not only to identify vulnerabilities but also to create working exploits in controlled environments. However, it is worth noting that GPT-5.5-Cyber outperformed Anthropic Mythos 5 on CyberGym by only 1.8 percentage points (85.6% versus 83.8%), indicating ongoing competition in this niche.
A more complex picture emerges from data from the UK AI Security Institute. In an April report, AISI stated that GPT-5.5 completed a 32-step simulation of the "The Last Ones" attack in 2 out of 10 attempts, while Mythos Preview did so in 3 out of 10. However, in updated data, Mythos Preview improved its result to 6 out of 10, and GPT-5.5 to 3 out of 10. This suggests that, despite its high specialization, GPT-5.5-Cyber still lags in some scenarios involving complex multi-stage attacks.
Access Strategy and Partnerships
OpenAI emphasizes that GPT-5.5-Cyber is not intended for mass use. The model is designed for verified clients who require advanced cyber capabilities in authorized scenarios. For most users, the company recommends the standard GPT-5.5 with Trusted Access for Cyber and Codex Security. As part of the Daybreak Cyber Partner Program, OpenAI has already attracted giants such as Akamai, Cisco, CrowdStrike, IBM, and Palo Alto Networks.
Particularly noteworthy is the Patch the Planet initiative, aimed at supporting open-source projects. In collaboration with Trail of Bits, HackerOne, and other participants, the program has already attracted projects such as cURL, Go, Python, and Sigstore. Since March, the cloud version of Codex Security has scanned over 30 million commits across 30,000 code bases, with human reviewers noting more than 70,000 findings as fixed.
Expert Analysis
The launch of GPT-5.5-Cyber against the backdrop of Anthropic's restrictions is not just a technological step but a clear signal to the market. OpenAI demonstrates a willingness to operate within regulatory requirements without sacrificing innovation. However, AISI data shows that the arms race in AI cybersecurity is far from over. The gap between models is narrowing, and in the coming months, we will see even more aggressive competition. Investors and developers should closely monitor how these tools are integrated into real corporate systems — that is where their true value will be determined.