OpenAI has broadened access to its cybersecurity-focused AI model, GPT-5.4-Cyber, shortly after Anthropic introduced its highly restricted Mythos model, signaling intensifying competition in AI-driven cyber defense.
The model is being rolled out through OpenAI’s Trusted Access for Cyber (TAC) program, which is now expanding to include thousands of verified cybersecurity professionals and hundreds of security teams. This marks a shift toward wider—but still controlled—availability of advanced cyber capabilities, compared to Anthropic’s limited-access approach.
GPT-5.4-Cyber is specifically fine-tuned for defensive cybersecurity use cases and includes enhanced capabilities such as binary reverse engineering, allowing users to analyze software for vulnerabilities and malicious behavior. It also relaxes certain guardrails for legitimate security work, making it more effective for real-world threat analysis.
Access to the model is being granted in phases to vetted organizations, researchers, and security vendors, with individuals able to apply through a verification process. OpenAI emphasized three key principles behind the rollout: democratized access through objective verification, iterative deployment to refine the model based on real-world use, and strengthening ecosystem resilience by supporting the broader cybersecurity community.
The move comes in direct response to Anthropic’s Mythos model, which demonstrated the ability to autonomously discover and exploit thousands of software vulnerabilities. Due to its powerful and potentially risky capabilities, Mythos has been restricted to a small group of major organizations under a tightly controlled program.
