OpenAI releases GPT‑5.4‑Cyber

OpenAI has released GPT‑5.4‑Cyber, a version of its flagship model tuned for defensive cybersecurity and limited to vetted users. (openai.com) OpenAI said on April 14 that it is expanding its Trusted Access for Cyber program to “thousands” of verified individual defenders and “hundreds” of teams that secure critical software. The company said GPT‑5.4‑Cyber is the first model in that rollout. (openai.com) Cybersecurity work means finding weaknesses before attackers do, checking suspicious code, and tracing how malicious software behaves. OpenAI said it fine-tuned GPT‑5.4‑Cyber to be “cyber-permissive,” meaning it is more willing than a general model to help with defensive tasks that can look dangerous out of context. (openai.com) That change follows a broader shift among artificial intelligence companies toward restricted access for high-risk cyber tools. OpenAI launched Trusted Access for Cyber in February 2026 as an identity-based program and committed $10 million in application programming interface credits for cyber defense work. (openai.com) GPT‑5.4‑Cyber is built on GPT‑5.4, which OpenAI introduced on March 5 as its main professional model with a 1 million-token context window, stronger coding performance, and native computer-use features in Codex and the application programming interface. Those base capabilities make it more useful for long code reviews and multi-step security investigations. (openai.com) Early coverage said the new model can help reverse engineer binaries, which means taking compiled software apart to understand what it does when the source code is not available. XDA Developers reported that capability after OpenAI opened testing to approved security users rather than the general public. (xda-developers.com) Reuters reported the launch came one week after Anthropic announced Mythos, its own cyber-focused model, as competition between model makers moved deeper into security work. The Hacker News also said OpenAI framed the release as part of a push to help defenders find and fix problems faster. (usnews.com) (thehackernews.com) OpenAI said access is still gated by verification and safeguards, not opened broadly across ChatGPT. The company said it is preparing for “increasingly more capable models” over the next few months, with GPT‑5.4‑Cyber serving as an early test of how far it can widen access without dropping identity checks. (openai.com)