OpenAI's GPT‑5.4‑Cyber

OpenAI on April 14 began rolling out GPT‑5.4‑Cyber, a version of its latest model tuned for defensive cybersecurity and limited to vetted users. (openai.com) The company said the model will go first to vetted security vendors, organizations, and researchers, not to the general ChatGPT audience. Reuters reported the launch came one week after Anthropic announced Claude Mythos for similar defensive use under a controlled program. (openai.com) (reuters.com) Cybersecurity work here means using artificial intelligence to inspect software the way a mechanic inspects an engine: looking for flaws, weak points, and signs of tampering before attackers exploit them. OpenAI said GPT‑5.4‑Cyber lowers refusal barriers for legitimate security tasks and can help with binary reverse engineering, which means examining compiled software when the original source code is unavailable. (livemint.com) (openai.com) The release extends OpenAI’s Trusted Access for Cyber program, which the company introduced on February 5, 2026, as an identity-and-trust gate for higher-risk cyber uses. OpenAI said it is now scaling that program to thousands of verified individual defenders and hundreds of teams that protect critical software. (openai.com 1) (openai.com 2) OpenAI has been tightening cyber controls as its models become better at coding and autonomous computer tasks. When GPT‑5.4 launched on March 5, the company said it was its first general-purpose model with native computer-use features and up to 1 million tokens of context, while its system card said GPT‑5.4 Thinking was the first general-purpose model in the series with mitigations for “High” cybersecurity capability. (openai.com 1) (openai.com 2) That matters because the same tools that help defenders patch systems can also help attackers probe them. OpenAI said its broader cyber safety setup includes training models to refuse clearly malicious requests, automated monitors for suspicious activity, and rerouting some risky traffic to a less cyber-capable model. (openai.com 1) (openai.com 2) OpenAI said it has supported cyber defense work since 2023 through its Cybersecurity Grant Program and, in February, attached $10 million in application programming interface credits to the Trusted Access for Cyber effort. The company has framed the new rollout as preparation for “increasingly more capable models” expected over the next few months. (openai.com) (openai.com) (livemint.com) Anthropic is moving on a parallel track. Reuters said its Claude Mythos preview, announced April 7 under Project Glasswing, is also being kept to select organizations for defensive use and has already found “thousands” of major vulnerabilities in operating systems, web browsers, and other software. (reuters.com) For now, OpenAI is drawing a bright line between public chatbot access and security work that requires identity checks. The company said users can apply through its cyber access flow, while enterprise teams must go through OpenAI representatives. (openai.com) (openai.com)