OpenAI unveils cyber model
OpenAI introduced a cyber‑focused GPT‑5.4 model aimed at defensive security tasks and restricted to vetted vendors, researchers and organisations. The company says the model won’t appear on public ChatGPT and is designed for cybersecurity professionals rather than general users. (techradar.com) (livemint.com)
OpenAI has started rolling out GPT-5.4-Cyber, a security-focused version of GPT-5.4 for vetted defenders rather than public ChatGPT users. (openai.com) The company said on April 14, 2026 that the model is being offered through its Trusted Access for Cyber program, which it is expanding to thousands of verified individual defenders and hundreds of teams that protect critical software. (openai.com) OpenAI described GPT-5.4-Cyber as a “cyber-permissive” variant fine-tuned for defensive cybersecurity work, a category that includes finding vulnerabilities, analyzing malware, and helping security teams respond faster. (openai.com) Cybersecurity models sit in a narrow lane because the same skills used to patch a system can also be used to break into one. OpenAI said that ambiguity is why it built Trusted Access for Cyber as an identity- and trust-based gate instead of releasing the model broadly. (openai.com) The launch follows a broader shift inside OpenAI’s product line. When GPT-5.4 was released on March 5, 2026, the company put the general model in ChatGPT, the application programming interface, and Codex, while keeping the cyber-specific version separate. (openai.com) OpenAI has been moving toward tighter cyber controls for months as its models improved on security tasks. In a December 10, 2025 safety post, the company said capture-the-flag performance had risen from 27% on GPT-5 in August 2025 to 76% on GPT-5.1-Codex-Max in November 2025. (openai.com) That earlier post also said OpenAI was preparing for models that could reach “High” cybersecurity capability under its Preparedness Framework, including systems that could help develop zero-day exploits or assist with stealthy intrusions. (openai.com) OpenAI’s February 5, 2026 launch of Trusted Access for Cyber previewed this approach. The company said it would combine model refusals for clearly malicious requests with automated monitors for suspicious cyber activity and $10 million in application programming interface credits for defense work. (openai.com) The new release puts that policy into a product: a stronger cyber model for security professionals, with identity checks and limited distribution built in from the start. OpenAI said it expects to keep adjusting access and safeguards as more capable models arrive in the next few months. (openai.com)
