OpenAI releases GPT‑5.4‑Cyber

OpenAI has begun giving vetted security professionals access to GPT-5.4-Cyber, a new model tuned for defensive cyber work rather than general chat. (openai.com) The rollout started Tuesday, April 14, through OpenAI’s Trusted Access for Cyber program, not through standard ChatGPT. OpenAI said it is expanding that program to thousands of verified individual defenders and hundreds of teams that protect critical software. (openai.com) Cybersecurity work often means inspecting malicious code, tracing how an attack moves, or unpacking a compiled program when the original source code is missing. OpenAI said GPT-5.4-Cyber is fine-tuned to be more permissive on those defensive tasks for approved users, including reverse engineering binaries. (openai.com) The release comes a week after Anthropic announced Mythos, another frontier model aimed at cyber use, and Reuters reported OpenAI’s launch as part of a fast-moving contest over how much access advanced models should get in security work. Axios reported OpenAI is pairing broader access with tiered controls on who can use the model. (reuters.com) (axios.com) OpenAI’s bet is that specialized access programs can put stronger tools in defenders’ hands without releasing the same capability to the public internet. In its April 14 post, the company said it is preparing for “increasingly more capable models” over the next few months and is adjusting safeguards around cyber use now. (openai.com) GPT-5.4-Cyber builds on GPT-5.4, OpenAI’s flagship model released March 5 for professional work in ChatGPT, the application programming interface, and Codex. OpenAI described the base model as its most capable and efficient frontier model for professional tasks, with coding, tool use, and a context window of up to 1 million tokens. (openai.com) (developers.openai.com) OpenAI said access will be granted gradually to security vendors, researchers, nonprofit groups, and verified organizations, with identity checks and use restrictions tied to the program. Axios reported the company’s plan creates different access tiers rather than a single public release. (openai.com) (axios.com) The company is also framing the release as part of a broader cyber-defense ecosystem, not just a model launch. OpenAI said the program includes safeguards, monitoring, and support for defenders responsible for securing widely used software and infrastructure. (openai.com) For now, the practical shift is simple: OpenAI is moving advanced cyber help behind a vetting gate, and the first people through it are security teams, not ordinary chatbot users. (openai.com)