OpenAI releases GPT‑5.4‑Cyber

OpenAI has released GPT‑5.4‑Cyber, a version of its GPT‑5.4 model tuned for defensive security work and limited to vetted users. (openai.com) The model is being distributed through OpenAI’s Trusted Access for Cyber program, which the company said on April 14, 2026 it is expanding to thousands of verified individual defenders and hundreds of teams that protect critical software. (openai.com) Cybersecurity work often means finding weak spots before attackers do, then explaining how to fix them. OpenAI said GPT‑5.4‑Cyber is “cyber‑permissive,” meaning it is tuned to better handle defensive tasks such as vulnerability research, incident analysis, and binary reverse engineering, which is the process of inspecting compiled software when source code is unavailable. (openai.com) OpenAI is not offering the model as a public chatbot. The company said access is tied to identity checks, trust reviews, usage policies, and monitoring intended to keep higher-risk cyber capabilities in the hands of defenders rather than general users. (openai.com) The release comes as artificial intelligence companies are building more specialized security models instead of relying only on general-purpose assistants. CyberScoop reported OpenAI’s move puts it in more direct competition with Anthropic’s restricted cyber offerings, including Project Glasswing, over which companies and researchers get access to frontier security tools. (cyberscoop.com) OpenAI framed the timing around stronger models arriving soon. In its April 14 post, the company said it was preparing for “increasingly more capable models” over the next few months and was fine-tuning systems now so defenders could use them before attackers adapt. (openai.com) The company has been laying the policy groundwork for months. In February, OpenAI introduced Trusted Access for Cyber as an identity- and trust-based framework and said it would commit $10 million in application programming interface credits to support defensive work. (openai.com) This also builds on safeguards OpenAI had already added to its broader GPT‑5.4 line. In the GPT‑5.4 Thinking system card published on March 5, 2026, OpenAI said that model was its first general-purpose system with mitigations for “High” cybersecurity capability, a sign that cyber misuse had become a first-order safety concern inside the company’s release process. (openai.com) OpenAI said the aim is faster defense, not wider access. For now, GPT‑5.4‑Cyber is being rolled out through a gatekeeping program that treats advanced security help less like a mass-market product and more like controlled lab equipment. (openai.com)