OpenAI Launches GPT-5.4-Cyber for Defense Teams

OpenAI released GPT-5.4-Cyber, a fine-tuned model with lowered refusal boundaries for legitimate security work, including binary reverse engineering. Access is gated through the Trusted Access for Cyber program, initially limited to vetted security vendors and researchers before scaling to thousands of defenders. OpenAI is also committing $10 million in API credits through its Cybersecurity Grant Program, targeting teams with track records in open source and critical infrastructure vulnerability work. The timing is hard to miss – Anthropic recently launched its own cybersecurity-focused model, and the two companies are now directly competing for the trust of security teams. Restricted-access cybersecurity models are becoming a distinct product category, fast.