OpenAI’s GPT‑5.4‑Cyber rollout

OpenAI has started rolling out GPT-5.4-Cyber, a version of its flagship model tuned for defensive cybersecurity and limited to vetted users. (openai.com) The company said on April 14 that the model is being released first to vetted security vendors, organizations and researchers, not the general public. Reuters reported the launch came one week after Anthropic announced its own cyber model, Mythos, on April 7. (openai.com) (reuters.com) OpenAI is also expanding its Trusted Access for Cyber program to thousands of verified individual defenders and hundreds of teams protecting critical software. Reuters reported the program now has added tiers, with higher levels of verification unlocking more powerful capabilities, including GPT-5.4-Cyber at the top tier. (openai.com) (reuters.com) Cybersecurity work often involves dual-use tasks: the same steps used to find a software flaw can also be used to exploit it. Axios reported OpenAI’s new access plan is built around that problem, expanding access to advanced cyber models while tightening controls over who can use them. (axios.com) OpenAI said GPT-5.4-Cyber is “cyber-permissive,” meaning it is tuned to refuse less often on legitimate security tasks that standard models may block. The company said it is preparing for “increasingly more capable models” in the next few months and wants cyber defenses to scale in step. (openai.com) The base model underneath this release is GPT-5.4, which OpenAI introduced on March 5 for ChatGPT, the application programming interface, and Codex. OpenAI said GPT-5.4 supports up to 1 million tokens of context and is its first general-purpose model with native computer-use capabilities in Codex and the application programming interface. (openai.com) Trusted Access for Cyber itself is new. OpenAI launched the program on February 5 as an identity- and trust-based framework for giving defenders broader access to frontier cyber capabilities while trying to reduce misuse, and paired it with $10 million in application programming interface credits for defensive work. (openai.com) OpenAI says it has been building this cyber effort since 2023 through a Cybersecurity Grant Program, cyber evaluations, and a Preparedness Framework for model risks. In March, it also introduced Codex Security, an application security agent that analyzes codebases for vulnerabilities and proposes fixes. (openai.com 1) (openai.com 2) On April 14, OpenAI said Codex Security had contributed to fixes for more than 3,000 critical and high-severity vulnerabilities since its recent broader launch. The company’s pitch is that stronger models will be used by attackers and defenders alike, so access, identity checks and safeguards now move together. (openai.com)