On Tuesday, OpenAI announced the launch of “GPT-5.4-Cyber”, an improved version of its latest flagship model, “GPT-5.4”, developed specifically to meet the needs of defensive cybersecurity, just days after rival company “Anthropic” announced its flagship model, “Mythos”.
“OpenAI” indicated that the gradual use of artificial intelligence accelerates the work of defenders responsible for protecting systems, data, and users, enabling them to detect and address problems more quickly within the digital infrastructure.
At the same time, the company announced the expansion of its Trusted Access for Cybersecurity program to include thousands of authenticated individual defenders and hundreds of teams responsible for securing critical software.
The company explained that artificial intelligence systems remain dual-use in nature, as attackers can exploit technologies developed for legitimate purposes to achieve malicious goals, and among the main concerns is the use of custom software protection models to discover and exploit vulnerabilities within widespread programs before fixing them.
“OpenAI” confirmed that its goal is to expand the availability of its models while reducing their misuse, through a gradual and thoughtful rollout that gives defenders the advantage, in parallel with enhancing protection against system penetration and malicious injection attacks as the models’ capabilities develop.
The company added: “As the model’s capabilities evolve, our approach is to simultaneously expand the scope of cyber defense, by expanding access to legitimate defenders and continuing to strengthen safeguards.”
The company that developed “ChatGPT” also revealed that “Codex Security”, which was launched to find and verify security vulnerabilities and suggest solutions for them, has contributed to fixing more than 3,000 serious and high-risk security vulnerabilities.
This limited launch comes after a preview of Anthropic’s “Mythos” model, which is being deployed in a controlled manner within the “Glasswing” project, after the company announced that it was able to discover “thousands” of security vulnerabilities in operating systems, web browsers, and other programs.
“OpenAI” confirmed that the most powerful system is the one that constantly monitors, verifies, and addresses security problems while writing software, considering that integrating advanced programming models and intelligent capabilities into developers’ workflow provides immediate and actionable feedback, and moves security from mere periodic audits to a continuous and tangible reduction of risks.
