OpenAI has unveiled its specialized model GPT-5.4-Cyber

OpenAI has unveiled its specialized model GPT-5.4-Cyber

Since 2023, OpenAI has been awarding grants to security-focused participants and developing dedicated models aimed at identifying cybersecurity threats.

Introduced in March 2026, the Codex Security model, designed to work with application source code and help developers implement secure coding practices, had already been available in closed access back in 2025.

The Trusted Access for Cyber (TAC) program was announced in February 2026, and it already brings together hundreds of companies and thousands of security professionals. Joining the program requires a multi‑stage verification process, with each level unlocking additional capabilities for using OpenAI products.

Now, within the program, customers with a high TAC level have access to the GPT-5.4-Cyber model. This model is optimized for vulnerability analysis and has a lower refusal rate, which gives it enhanced capabilities in both offensive cybersecurity and reverse engineering, including malware analysis. One of its key features is the ability to study binary files without access to source code.

Open AI GPT-5.4-Cyber is available to a broader range of participants than Anthropic's Mythos model, and its user base is expected to grow under the current distribution policy.

💬 Discuss