OpenAI is preparing to launch GPT - 5.5-Cyber , a specialized cybersecurity AI model that will be made available exclusively to select “ critical cyber defenders .” CEO Sam Altman announced that the model will not be available to the general public, but will be distributed to trusted security professionals to strengthen organizations’ cyber defenses.
See also: OpenAI GPT-5.4-Cyber: New AI model for cybersecurity
The limited release is expected to happen “ in the coming days ,” Altman told X. “ We will work with the entire ecosystem and the government to determine trusted access for Cyber ,” he said. This approach reflects a growing trend in the AI industry , where companies are labeling their leading models as too risky for public release due to the potential for misuse.
GPT -5.5-Cyber is based on the recently released GPT-5.5, which OpenAI described asits “most intelligent and intuitive model to date.” While no technical details have been revealed, the name suggests a specialized version with enhanced cybersecurity capabilities. According to the Preparedness Framework , the model has been categorized as “High” in cybersecurity, as it can help identify vulnerabilities and cyber operations without reaching the “Critical” level.
GPT-5.5-Cyber evaluation and capabilities
The new model has outperformed all previous OpenAI in critical cybersecurity benchmarks. Specifically, it shows excellent performance in CTF challenges (professional vulnerability detection and exploitation), CVE-Bench (web application vulnerabilities), Cyber Range (end-to-end cyber operations), and VulnLMP (real-world exploits). Evaluations by the US CAISI and the UK AI Safety Institute confirmed the model’s improved capabilities.
Researchers describe GPT-5.5-Cyber as a “competent junior researcher” for vulnerability work, but not fully autonomous. Its specialized knowledge surpasses some expert operators, potentially automating bottlenecks in vulnerability discovery and exploitation. However, it retains significant limitations that prevent full autonomy in critical cyber operations.
See also: OpenAI announces new, more efficient GPT-5.5 model
OpenAI has implemented advanced security systems, including automated classifiers that detect suspicious activity and direct it to less capable models. These safeguards have been improved upon from GPT-5.2 ’s controls and include restrictions on high-risk workflows, activity monitoring, and authenticated controls .
Competition and market strategy
OpenAI 's move comes in direct contrast to Anthropic , which is keeping its Claude Mythos model off the public due to security concerns. OpenAI President Greg Brockman highlighted GPT-5.5 's improvements in coding, computer use, agentic tasks, and cybersecurity, highlighting the public release as opposed to Anthropic 's approach .
The White House has shown strong interest in the Mythos, despite ongoing tensions with Anthropic following its disputes with the Pentagon. Reports indicate that the White House has opposed plans to expand access to Mythos, citing cybersecurity concerns and fears that increased demand could hinder the government's ability to effectively use the system.
The trend toward gradual releases reflects a growing awareness of the risks associated with advanced AI models. OpenAI has previously implemented similar strategies for cybersecurity-focused models, as well as for its new life sciences model GPT-Rosalind, designed to support biological research and drug discovery.
See also: OpenAI Codex Security: Identified 10,561 serious issues in 1.2 million commits
Impacts and future prospects
For cyber defenders, GPT-5.5-Cyber promises to alleviate scaling bottlenecks by automating vulnerability discovery and exploitation, acting as a skilled assistant for security researchers. Companies and organizations will gain enhanced defenses through trusted access, with security measures limiting misuse while preserving beneficial uses.
