OpenAI Begins GPT-5.5-Cyber Rollout to Critical Defenders

OpenAI Unveils Specialised Cyber Defence Model

OpenAI CEO Sam Altman announced on Wednesday the rollout of GPT-5.5-Cyber, a frontier AI model built specifically for cybersecurity applications, pledging to deliver it to critical cyber defenders within days. The move marks a significant escalation in the race to deploy AI-powered security tools, coming weeks after rival Anthropic unveiled Claude Mythos Preview, a model deemed too dangerous to release publicly due to its vulnerability-discovery capabilities.

A Broader Distribution Strategy

The announcement accompanied a broader cybersecurity action plan published by OpenAI, structured around five pillars: democratising access to cyber defence tools, coordinating between government and industry, strengthening safeguards around advanced capabilities, maintaining deployment visibility, and enabling users to protect themselves. GPT-5.5-Cyber builds on GPT-5.4-Cyber, which OpenAI introduced in mid-April through its Trusted Access for Cyber programme alongside $10 million in API grants for vetted security organisations.

Where Anthropic restricted Mythos to roughly 50 organisations under the tightly controlled Project Glasswing consortium, OpenAI is positioning its approach as significantly more expansive. GPT-5.5-Cyber will be distributed through the TAC programme to governments, critical infrastructure operators, security vendors, cloud platforms, and financial institutions.

The Mythos Shadow Looms Large

Anthropic's Claude Mythos Preview, announced on 7th April, cast a long shadow over the industry. The model autonomously discovered thousands of previously unknown zero-day vulnerabilities across every major operating system and web browser. In one demonstration, Mythos constructed a full browser exploit chaining four separate vulnerabilities to escape both renderer and operating system sandboxes; in another, it developed a FreeBSD remote code execution exploit requiring a 20-gadget ROP chain.

Anthropic committed up to $100 million in usage credits and $4 million in donations to open-source security organisations, but was unequivocal that Mythos would not be made publicly available.

Benchmarks Remain Unclear

OpenAI has not published technical benchmarks comparing GPT-5.5-Cyber to Mythos. The base GPT-5.5 model, released on 23rd April, was rated High on OpenAI's own cybersecurity risk scale, below the Critical threshold that would indicate autonomous zero-day exploit development. Whether the Cyber-specific variant closes that gap remains unclear, but the consensus among security professionals is that the capability to find software flaws at scale is no longer containable.