What is the story about?
OpenAI is taking a more direct step into the cybersecurity race and Sam Altman appears eager to move quickly. The AI firm has finally launched it GPT-5.5-Cyber model for limited preview.
Launching the new model, Altman wrote, “We'd like to help companies secure themselves and we think it's important to start work on this quickly,” Altman wrote in a post on X.
The comment accompanied the announcement of GPT-5.5-Cyber, a specialised variation of the company’s recently launched GPT-5.5 model that is now rolling out in limited preview to carefully vetted security teams.
The timing is notable. Just a month ago, rival AI company Anthropic drew widespread attention from investors, policymakers and security circles with the debut of
Claude Mythos Preview
, its own cyber-focused AI system.
Now, OpenAI is signalling that it does not intend to sit on the sidelines as AI companies increasingly position themselves as key players in digital defence infrastructure.
Unlike a traditional next-generation model launch, OpenAI says GPT-5.5-Cyber is not primarily about dramatically increasing cyber capabilities. Instead, the focus is on making the model more permissive and useful for legitimate security workflows that would otherwise face stricter safeguards in the standard version of GPT-5.5.
According to the company, vetted cybersecurity professionals will be able to use the model more effectively for tasks such as vulnerability identification, triage operations, patch validation and malware analysis. These are areas where conventional AI safety restrictions can sometimes limit deeper technical investigation.
OpenAI described the release as an opportunity for a smaller set of partners to explore “advanced workflows where specialised access behaviour may matter”.
The company has been careful to stress that access is restricted and monitored. The preview is being positioned as a controlled environment for cybersecurity defenders, particularly those working to secure critical infrastructure and enterprise systems.
The move also reflects a growing shift in the AI industry. Rather than treating cybersecurity as a secondary use case, major AI labs are increasingly building dedicated models and access programmes specifically for defence-oriented work.
Anthropic triggered similar conversations across Washington and Silicon Valley last month when it unveiled Claude Mythos Preview under a new initiative called Project Glasswing.
The company limited access to select organisations and reportedly held discussions with senior US government officials ahead of the rollout. Anthropic CEO Dario Amodei met members of the Trump administration to discuss the model’s cyber capabilities and broader implications.
The launch also sparked interest among financial institutions and policymakers. Reports suggested that Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent met with major US banking executives to discuss the emerging risks and opportunities surrounding advanced cybersecurity AI systems.
Together, the launches from OpenAI and Anthropic suggest that AI companies are entering a new phase where securing digital infrastructure may become just as strategically important as building general-purpose chatbots.
Launching the new model, Altman wrote, “We'd like to help companies secure themselves and we think it's important to start work on this quickly,” Altman wrote in a post on X.
we'd like to help companies secure themselves and we think it's important to start work on this quickly https://t.co/dSobnv4SfX
— Sam Altman (@sama) May 8, 2026
The comment accompanied the announcement of GPT-5.5-Cyber, a specialised variation of the company’s recently launched GPT-5.5 model that is now rolling out in limited preview to carefully vetted security teams.
The timing is notable. Just a month ago, rival AI company Anthropic drew widespread attention from investors, policymakers and security circles with the debut of
Now, OpenAI is signalling that it does not intend to sit on the sidelines as AI companies increasingly position themselves as key players in digital defence infrastructure.
OpenAI rolls out GPT-5.5-Cyber in a limited preview
Unlike a traditional next-generation model launch, OpenAI says GPT-5.5-Cyber is not primarily about dramatically increasing cyber capabilities. Instead, the focus is on making the model more permissive and useful for legitimate security workflows that would otherwise face stricter safeguards in the standard version of GPT-5.5.
According to the company, vetted cybersecurity professionals will be able to use the model more effectively for tasks such as vulnerability identification, triage operations, patch validation and malware analysis. These are areas where conventional AI safety restrictions can sometimes limit deeper technical investigation.
OpenAI described the release as an opportunity for a smaller set of partners to explore “advanced workflows where specialised access behaviour may matter”.
The company has been careful to stress that access is restricted and monitored. The preview is being positioned as a controlled environment for cybersecurity defenders, particularly those working to secure critical infrastructure and enterprise systems.
The move also reflects a growing shift in the AI industry. Rather than treating cybersecurity as a secondary use case, major AI labs are increasingly building dedicated models and access programmes specifically for defence-oriented work.
Anthropic did the same last month
Anthropic triggered similar conversations across Washington and Silicon Valley last month when it unveiled Claude Mythos Preview under a new initiative called Project Glasswing.
The company limited access to select organisations and reportedly held discussions with senior US government officials ahead of the rollout. Anthropic CEO Dario Amodei met members of the Trump administration to discuss the model’s cyber capabilities and broader implications.
The launch also sparked interest among financial institutions and policymakers. Reports suggested that Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent met with major US banking executives to discuss the emerging risks and opportunities surrounding advanced cybersecurity AI systems.
Together, the launches from OpenAI and Anthropic suggest that AI companies are entering a new phase where securing digital infrastructure may become just as strategically important as building general-purpose chatbots.
