OpenAI has launched GPT-5.4-Cyber, a specialized version of its GPT-5.4 model, tailored for cybersecurity applications. This variant features a reduced refusal threshold for security-related tasks, allowing professionals to conduct operations like binary reverse engineering, which standard models typically reject. Access to GPT-5.4-Cyber is restricted to vetted security vendors, enterprise security teams, and researchers through OpenAI's Trusted Access for Cyber (TAC) program. The TAC program, which now includes thousands of defenders and hundreds of security teams, offers tiered access to the model. Individual users can verify their identity for basic access, while enterprises must apply through an OpenAI representative. OpenAI's tiered strategy aims to balance safety with specialized capabilities, ensuring that permissive models are deployed securely while standard models remain broadly accessible.