OpenAI launches GPT‑5.4‑Cyber

OpenAI has started rolling out GPT‑5.4‑Cyber, a cybersecurity version of its GPT‑5.4 model, to vetted defenders through a gated access program. (openai.com) OpenAI said on April 14, 2026 that it is expanding Trusted Access for Cyber to thousands of verified individual defenders and hundreds of teams responsible for defending critical software. The company said GPT‑5.4‑Cyber is the first model in that push. (openai.com) The model is tuned for jobs security teams already do by hand: reading unfamiliar code, analyzing software flaws, and examining malware, which is malicious software used to break into or damage systems. OpenAI said the model is “cyber-permissive,” meaning it is trained to be more willing to help with defensive security tasks than its general-purpose models. (openai.com) (helpnetsecurity.com) Cybersecurity is the work of finding weak spots before attackers do, then fixing them fast. OpenAI’s move comes as model makers are building systems that can both help defenders audit software and, if misused, help attackers search for bugs at greater speed. (openai.com) (reuters.com) OpenAI is making access identity-based instead of public. The company said vetted users can include independent researchers, enterprise security teams, and organizations defending critical infrastructure, while higher-risk users are screened out through the Trusted Access for Cyber process. (openai.com) (cyberscoop.com) The launch also lands in the middle of a competition over who can ship the most capable cyber model without opening it to abuse. Reuters reported OpenAI announced GPT‑5.4‑Cyber about a week after Anthropic introduced Mythos, another restricted cybersecurity model. (reuters.com) (thenextweb.com) OpenAI said it is pairing the rollout with outside testing and ecosystem support. The company said the United States Center for AI Standards and Innovation and the United Kingdom AI Security Institute received access for evaluations, and it named Bank of America, Cisco, CrowdStrike, Goldman Sachs, NVIDIA, Oracle, SpecterOps, and Zscaler among early participants. (openai.com) This is not OpenAI’s first cyber gate, but it is a larger one. In February 2026, the company introduced Trusted Access for Cyber and committed $10 million in application programming interface credits to cyber defense work; in April it widened that program and attached a dedicated GPT‑5.4 variant to it. (openai.com 1) (openai.com 2) OpenAI said more capable cyber systems are coming “over the next few months,” and framed GPT‑5.4‑Cyber as an early step in that rollout. For now, the company is betting that tighter vetting, limited distribution, and outside evaluations will let defenders use stronger tools before attackers do. (openai.com)