OpenAI is launching GPT-5.4-Cyber, a model tailored for defensive cybersecurity, designed to answer sensitive queries for verified security professionals. It includes binary reverse engineering capabilities. This launch coincides with an expansion of OpenAI’s Trusted Access for Cyber program, which offers access to thousands of verified users. The program requires users to authenticate and may involve waiving Zero-Data Retention for the highest-tier access. OpenAI’s approach contrasts with Anthropic’s restricted access to its Mythos model, emphasizing broad, verified access versus tightly controlled deployment. OpenAI’s strategy centers on access control, verification, and monitoring for security, while Anthropic restricts more powerful tools to select organizations. GPT-5.4-Cyber aims to support defensive security professionals with features like binary reverse engineering and handling dual-use queries. OpenAI’s Codex Security tool complements this model, scanning open-source projects for vulnerabilities. The cybersecurity AI landscape is diverging, with OpenAI advocating broader access and Anthropic emphasizing caution. The EU AI Act may impact how these models are governed, but for now, both companies are focused on enhancing cybersecurity capabilities in unprecedented ways.
OpenAI Launches Cybersecurity Model to Thousands of Defenders in Race with Anthropic’s Mythos
1 Min Read