OpenAI launches GPT 5.4 Cyber to boost AI-powered cybersecurity, eyes lead over Claude Mythos

On March 14, OpenAI revealed an expansion of its Trusted Access for Cyber (TAC) program with the introduction of the GPT 5.4 Cyber model. This innovative variant of GPT-5.4 is specifically engineered to enhance AI-driven cybersecurity measures, equipping security experts with the tools needed for sophisticated defensive tasks. This announcement follows closely on the heels of Anthropic's launch of Claude Mythos as part of Project Glasswing, highlighting the escalating competition among tech giants to create advanced tools capable of identifying vulnerabilities and combating new threats in the digital arena. The standout feature of GPT 5.4 Cyber lies in its binary reverse engineering capability. This allows cybersecurity professionals to analyze software, uncover malware threats, identify security flaws, and evaluate the robustness of applications—all without requiring access to the source code. The GPT 5.4 Cyber model is fine-tuned specifically for cybersecurity applications and has fewer limitations compared to its predecessor. It is designed to facilitate complex defensive operations, including vulnerability detection in extensive codebases, reasoning through intricate software frameworks, and analyzing malware behaviors. Due to its enhanced capabilities, OpenAI is proceeding with caution regarding the model's distribution. Access to GPT 5.4 Cyber is limited to vetted security firms, organizations, and researchers. OpenAI emphasized that the use of these advanced cyber-capable models will come with restrictions, especially concerning no-visibility scenarios like Zero-Data Retention (ZDR). When comparing it to Anthropic’s Claude Mythos, it's important to note that while both models target cybersecurity, GPT 5.4 Cyber is an adapted version of an existing model, whereas Claude Mythos is a completely new creation. Currently, Mythos is available only to a select group of organizations under Project Glasswing for evaluation, whereas GPT 5.4 Cyber is accessible through a controlled TAC program, which was initially introduced in February 2026, allowing for stringent identity verification before access is granted.