OpenAI launches GPT-5.5-Cyber for vetted security teams in limited preview

OpenAI has announced the limited preview rollout of GPT-5.5-Cyber, a specialised variation of its latest artificial intelligence model designed for vetted cybersecurity teams.

The announcement, made on Thursday, comes exactly one month after rival firm Anthropic captured the attention of government officials with the release of Claude Mythos Preview.

According to a formal blog post, OpenAI explained that this specific version is not intended to be a major leap in raw capability, but is instead trained to be more "permissive" regarding security-related tasks.

The cyber-specific model is engineered to facilitate workflows such as vulnerability identification, triage, patch validation, and malware analysis.

OpenAI noted that the standard safeguards built into the generally available GPT-5.5 model, released late last month, often made these legitimate security tasks challenging for professionals.

"GPT-5.5-Cyber lets a smaller set of partners study advanced workflows where specialized access behavior may matter," the company stated.

This controlled release strategy mirrors Anthropic’s Project Glasswing, which also restricted access to a select group of companies to manage the model's potential power.

The move follows high-level discussions between tech leaders and government officials, including Treasury Secretary Scott Bessent and Vice President JD Vance, regarding the implications of advanced AI in national security.

Anthropic CEO Dario Amodei recently met with members of the Trump administration to discuss these technologies, despite previous regulatory friction.

While GPT-5.5-Cyber remains in a restricted phase, its deployment marks a significant step in the ongoing race to provide defensive tools against evolving digital threats. OpenAI intends to monitor performance closely before considering a wider release.