OpenAI rolls out GPT‑5.4‑Cyber
OpenAI released GPT‑5.4‑Cyber, a version of its flagship model fine-tuned for defensive cybersecurity work and made available only to verified security professionals through a Trusted Access for Cyber programme. The launch mirrors a recent pattern of gated specialist models aimed at sensitive use cases rather than broad public access. (reuters.com, siliconangle.com)
OpenAI has started rolling out GPT-5.4-Cyber, a cybersecurity version of its flagship model, only to vetted defenders and researchers. (reuters.com) The company announced the model on Tuesday, April 14, and said access will begin on a limited basis for security vendors, organizations, and researchers because the system is more permissive on sensitive cyber tasks. (reuters.com) Cybersecurity models are built to help find software flaws, analyze attacks, and speed up patching, but the same skills can also be misused. OpenAI said approved users at the highest verification tier will get fewer restrictions on work such as vulnerability research and analysis. (openai.com, reuters.com) OpenAI launched its Trusted Access for Cyber program on February 5, 2026 as an identity- and trust-based pilot for defensive work, and said it was committing $10 million in application programming interface credits to cyber defense. On April 14, it said the program was being expanded to thousands of verified individual defenders and hundreds of teams protecting critical software. (openai.com, reuters.com) The release follows a pattern in which frontier artificial intelligence companies keep their most sensitive cyber systems behind gates instead of putting them into broad public products. Reuters reported that Anthropic announced its own model, Mythos, on April 7 through a controlled program called Project Glasswing. (reuters.com) That caution builds on OpenAI’s earlier safety work around GPT-5.4. In its March 5 system card, the company said GPT-5.4 was its first general-purpose model with mitigations for what it called “High capability in Cybersecurity.” (openai.com, openai.com) GPT-5.4 itself was released on March 5 for ChatGPT, the application programming interface, and Codex, with up to 1 million tokens of context and native computer-use features for software and workflow tasks. GPT-5.4-Cyber takes that base model and narrows access rather than widening it. (openai.com, reuters.com) OpenAI’s pitch is that stronger tools should reach defenders before attackers do. Its rollout suggests the company sees cyber models less as mass-market chatbots and more as controlled infrastructure for a small set of verified users. (openai.com, reuters.com)
