OpenAI launches GPT‑5.4‑Cyber

OpenAI has released GPT‑5.4‑Cyber, a version of its flagship model tuned for defensive cybersecurity work and offered first to vetted users. (openai.com) The company announced the model on April 14, 2026, and said it is expanding its Trusted Access for Cyber program to thousands of verified individual defenders and hundreds of teams that protect critical software. (openai.com) GPT‑5.4‑Cyber is a fine-tuned variant of GPT‑5.4, which OpenAI introduced on March 5, 2026 as its flagship professional model with a 1 million-token context window, coding features, tool use, and computer-use capabilities. (openai.com) In plain terms, defensive cybersecurity means using a model to help find, analyze, and fix software flaws before attackers exploit them. OpenAI said this new version is trained to be more “cyber-permissive,” meaning it is more willing to answer legitimate security requests from approved defenders. (openai.com) That shift follows OpenAI’s earlier safety work around cyber-capable models. In its March 2026 system card for GPT‑5.4 Thinking, the company said GPT‑5.4 was its first general-purpose model with mitigations for “High” capability in cybersecurity. (openai.com) OpenAI is not making GPT‑5.4‑Cyber broadly available. Reuters reported the company plans a limited rollout to vetted security vendors, organizations, and researchers because the model has fewer restrictions on sensitive tasks such as vulnerability research and analysis. (reuters.com) The release landed one week after Anthropic announced Claude Mythos Preview through its Project Glasswing program. Anthropic said that effort is aimed at organizations responsible for critical software and listed partners including Amazon Web Services, Cisco, CrowdStrike, Google, Microsoft, NVIDIA, and Palo Alto Networks. (anthropic.com) That has turned cyber defense into a new front in the competition between frontier model companies. Reuters said OpenAI’s move came after Anthropic’s Mythos announcement, while Axios reported OpenAI paired the launch with a tiered access plan for advanced cyber models. (reuters.com) (axios.com) OpenAI has been building toward this for months. In February 2026, it launched Trusted Access for Cyber and committed $10 million in application programming interface credits to speed up defensive security work while tightening identity and trust checks around stronger cyber capabilities. (openai.com) The immediate test is whether vetted defenders use GPT‑5.4‑Cyber to shorten the time between finding a flaw and fixing it, without widening access to tools that could also help attackers. (openai.com)