OpenAI released GPT-5.4-Cyber on April 14, a fine-tuned variant of GPT-5.4 with lowered refusals for defensive cybersecurity tasks not available in the standard model, including binary reverse engineering of compiled software to surface malware and vulnerabilities without source code access. Access is gated through the expanded Trusted Access for Cyber program—individual defenders verify identity at chatgpt.com/cyber while enterprise security teams request access through OpenAI representatives. The rollout targets thousands of individual defenders and hundreds of security teams, with tiered verification levels controlling which capabilities are unlocked.