OpenAI's GPT‑5.4‑Cyber

OpenAI has introduced GPT‑5.4‑Cyber, a version of its flagship model reserved for vetted security users instead of the general public. (openai.com) The company announced the model on April 14, 2026 and said it is fine-tuned for defensive cybersecurity work such as finding and fixing software flaws faster. Reuters reported the launch came a week after Anthropic disclosed its own restricted cyber model, Mythos, on April 7. (openai.com) (reuters.com) Cybersecurity is the business of spotting weak points in software before criminals do. OpenAI said GPT‑5.4‑Cyber is trained to be more “cyber-permissive,” meaning it is less likely than a general chatbot to refuse legitimate security testing requests from approved users. (openai.com) OpenAI said access will run through its Trusted Access for Cyber program, which it is expanding to thousands of verified individual defenders and hundreds of teams responsible for protecting critical software. The company first introduced that trust-based access framework in February 2026. (openai.com 1) (openai.com 2) The restriction reflects a wider change in how artificial intelligence companies are handling models that can help with computer intrusion as well as defense. The New York Times reported OpenAI and Anthropic are both keeping these systems inside small partner groups rather than releasing them broadly. (nytimes.com) OpenAI tied the launch to a bigger jump in model capability. In its April 14 post, the company said it was preparing for “increasingly more capable models” over the next few months and was adjusting access controls and safeguards ahead of that shift. (openai.com) GPT‑5.4‑Cyber is built on GPT‑5.4, which OpenAI released on March 5, 2026 as its main professional model with stronger coding, tool use, computer control, and a context window of up to 1 million tokens. Those features make the base system more useful for software analysis, but they also raise the stakes for misuse. (openai.com) OpenAI said it believes the safeguards around GPT‑5.4‑Cyber are strong enough for restricted deployment. The company said those controls include identity checks, monitoring, and limits designed to place advanced cyber capabilities “in the right hands.” (openai.com 1) (openai.com 2) That leaves GPT‑5.4‑Cyber in a narrow lane: more willing to help defenders do real security work, but still distributed as a controlled tool rather than a public chatbot. (openai.com)