OpenAI, the developer of ChatGPT, has begun offering a new artificial intelligence (AI) model specialized for detecting security vulnerabilities to some corporations. It rolled out a similar security-dedicated model just a week after rival Anthropic unveiled "Mythos," known for its strong security performance, moving to respond.
OpenAI said on the 14th (local time) that it has released a new AI model, "GPT-5.4-Cyber," on a limited basis to some security experts and corporate customers. The security-specialized model is optimized to automatically find software security flaws (bugs).
The new model is available only to some participants in the "Trusted Access for Cyber" (TAC) program, which OpenAI launched in February. After testing the new model with hundreds of top-tier customers, OpenAI plans to expand access to thousands within a few weeks.
"GPT-5.4-Cyber" is a fine-tuned version of OpenAI's top-tier model GPT-5.4 for security tasks, equipped with "binary reverse engineering" capabilities that allow security experts to analyze software executables without source code to assess malware potential, vulnerabilities, and security hardening.
As AI corporations roll out models with strengths in detecting security vulnerabilities one after another, concerns are growing that hackers or criminal groups could exploit them. Hacker groups are already using AI to hack more easily than before, and there are worries that using cutting-edge AI could expand cyberattacks on all fronts in what is being called "Bugmageddon."
Earlier, right after Anthropic distributed "Claude Mythos" on a limited basis to some big tech corporations including Apple, Microsoft, and Amazon, the U.S. government and the financial sector convened an emergency meeting over Mythos' potential threats. The White House recently held an emergency review meeting with Treasury Secretary Scott Bessent, Federal Reserve (Fed) Chair Jerome Powell, and executives from major tech and financial industries to discuss countermeasures.