OpenAI restricts GPT-5.5-Cyber to select defenders weeks after criticizing Anthropic for similar approach

OpenAI is releasing GPT-5.5-Cyber to a limited group of "trusted defenders" in the coming days, CEO Sam Altman announced on X. Access will be restricted to handpicked cybersecurity professionals working to secure critical systems.

"We will work with the entire ecosystem and the government to figure out trusted access for cyber," Altman wrote, stating the goal is to "rapidly help secure companies and infrastructure."

The model's capabilities

According to OpenAI, GPT-5.5-Cyber can perform penetration testing, identify vulnerabilities, exploit them, and analyze malware. The UK's AI Security Institute stated this week that GPT-5.5-Cyber is "one of the strongest models we have tested on our cyber tasks." The institute noted it is only the second system to complete one of its multi-step attack simulations end to end.

Altman's previous criticism

The restricted release comes weeks after Anthropic deployed its cybersecurity-focused model Claude Mythos to approximately 50 organizations under strict controls. Anthropic stated Claude Mythos would never be made publicly available.

Altman criticized this approach during an appearance on the Core Memory podcast, as reported by TechCrunch. "There are people in the world who, for a long time, have wanted to keep AI in the hands of a smaller group of people," he said. "You can justify that in a lot of different ways."

He likened the strategy to selling fear: "We have built a bomb, we are about to drop it on your head. We will sell you a bomb shelter for $100 million."

Technical specifications

Pricing, context window size, parameter count, and other technical specifications have not been disclosed. OpenAI has not announced when or if GPT-5.5-Cyber will receive broader availability.

What this means

OpenAI is adopting the same gatekeeping strategy it recently mocked, signaling that both leading AI labs recognize the genuine dual-use risks of models with advanced cybersecurity capabilities. The restricted rollout suggests OpenAI has concluded that offensive security tools require careful distribution controls, regardless of past rhetoric about open access. Independent validation from the UK's AI Security Institute indicates the model's capabilities are substantive enough to warrant these concerns.