OpenAI Launches GPT-5.4-Cyber AI Model for Cybersecurity Applications

OpenAI has announced the limited release of its latest AI model, GPT-5.4-Cyber, designed for cybersecurity applications. The company is offering early access to select companies and users, including Amazon and Microsoft. OpenAI claims that GPT-5.4-Cyber can identify thousands of zero-day cybersecurity vulnerabilities, surpassing the capabilities of previous models.

Key Highlights

• OpenAI launches GPT-5.4-Cyber for cybersecurity in limited release to select companies
• Model identifies thousands of zero-day vulnerabilities and lowers refusal boundary for security research
• Access requires verification through Trusted Access for Cyber programme started in February 2026
• GPT-5.4-Cyber compared to Anthropic's Claude Mythos, both focus on cybersecurity applications

Key Features of GPT-5.4-Cyber

GPT-5.4-Cyber is a specialized version of the existing GPT-5.4 model. According to OpenAI, it lowers the refusal boundary for legitimate cybersecurity work. This means companies and researchers can use the model to detect cybersecurity flaws that standard GPT-5.4 may not identify. The model allows professionals to analyze software for malware potential and vulnerabilities, even without access to source code.

OpenAI is releasing GPT-5.4-Cyber through its Trusted Access for Cyber programme, which began in February 2026. The programme provides certain customers and cybersecurity professionals with access to OpenAI's most advanced tools. Users on the highest tier will receive access to the cyber model. Any user or company can request access after completing a verification process, which includes government ID authentication and additional steps. Companies may also request access through their OpenAI representative.

Comparison with Claude Mythos

GPT-5.4-Cyber is often compared to Anthropic's Claude Mythos, another AI model focused on cybersecurity. However, GPT-5.4-Cyber is a tailored version of an existing model, while Claude Mythos is a new generation not based on a previous version. Anthropic has made Claude Mythos available to around 40 companies through Project Glasswing. In contrast, OpenAI's model is part of the Trusted Access for Cyber programme.

The release of GPT-5.4-Cyber reflects a broader industry trend toward AI models dedicated to cybersecurity. Claude Mythos has already identified thousands of security flaws missed by human experts, raising concerns among financial institutions and government agencies. US Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell have led discussions about the implications of such technology. The Treasury Department is seeking access to Claude Mythos to help identify vulnerabilities.

Industry Impact and Access

The introduction of GPT-5.4-Cyber comes as AI models play a larger role in cybersecurity. OpenAI's approach allows for broader access, provided users complete verification. This move may influence how companies and agencies address cybersecurity threats in the future.