OpenAI Launches GPT-5.5-Cyber for Security Teams

OpenAI has released GPT-5.5-Cyber, a version of its latest AI model built specifically for cybersecurity professionals. The model launched on May 7 in limited preview and is only available to vetted defenders who protect critical infrastructure.

What Makes It Different

The standard version of GPT-5.5 refuses many requests related to security testing — asking it to analyse malware or write a proof-of-concept exploit typically triggers safety filters. GPT-5.5-Cyber removes those restrictions for authorised users, making it far more useful for real-world defensive security work.

Approved teams can use the model for vulnerability identification, malware analysis, binary reverse engineering, detection engineering, and patch validation. It can also help defenders write proof-of-concept exploits for bugs they discover and run simulations to test an organisation’s security posture.

Despite the loosened restrictions, OpenAI says the model still blocks genuinely malicious actions like credential theft, persistence techniques, and deploying malware against third-party systems.

The Bigger Picture

This release comes roughly a month after Anthropic launched Mythos Preview, its own cybersecurity-focused model. The two companies are taking different approaches to the same problem: Anthropic restricts Mythos access to about 40 organisations, while OpenAI is offering tiered access through its Trusted Access for Cyber programme.

In testing by the UK AI Security Institute, both models showed comparable results on expert-level cybersecurity tasks. GPT-5.5 achieved a 71.4% pass rate on expert tasks compared to 68.6% for Mythos Preview.

Starting June 1, all GPT-5.5-Cyber users will be required to enable Advanced Account Security on their OpenAI accounts.

What This Means for You

If you work in cybersecurity, this is a significant tool worth applying for. AI models that can reason about vulnerabilities, analyse malicious code, and help with penetration testing have been a gap in the market — standard chatbots are too cautious for real security work.

For everyone else, this signals a broader trend: AI companies are building specialised models for professional use cases rather than trying to make one model do everything. Expect more industry-specific AI tools in the months ahead.

If you are interested in how AI is changing security, check out our guide on getting started with AI tools and our newsletter for daily updates.