OpenAI Launches GPT-5.4-Cyber, Cybersecurity-Focused AI Model for Top-Tier Clients

Gate News message, April 15 — OpenAI has unveiled GPT-5.4-Cyber, a specialized cybersecurity AI model, to a limited group of top-tier clients under its trusted access program launched in February. The rollout comes one week after competitor Anthropic released Mythos, a similar model designed to identify and prevent software vulnerabilities.

GPT-5.4-Cyber is initially available only to the highest-tier users within OpenAI's trusted access program, currently numbering in the hundreds, with plans to expand to thousands within weeks. The model is designed to automatically detect software flaws and vulnerabilities, alerting security experts to address them before malicious actors can exploit them. OpenAI stated that Codex Security, its earlier cybersecurity model released in March, has already contributed to fixing over 3,000 critical and high-risk vulnerabilities.

The competitive push reflects growing concerns among governments and financial institutions about AI-enabled cyber threats. U.S. Vice President Vance recently held emergency discussions with major tech companies, while Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened leaders from major banks to address potential cyber risks posed by advanced AI models.