Artificial intelligence company OpenAI announced on Tuesday that it will release its latest cybersecurity-focused model to a limited group of partners, following a similar move by rival Anthropic, which also restricted access to a new system capable of identifying thousands of software vulnerabilities.
The cautious rollout by two leading AI firms highlights growing concerns about an emerging AI-driven arms race between cybersecurity defenders and malicious actors, who could exploit such tools for harmful purposes, News.Az reports, citing AFP.
“Our goal is to make these tools as widely available as possible while preventing misuse,” OpenAI said in a blog post.
Anthropic recently made its Claude Mythos model available to just 40 major technology organizations under an initiative known as Project Glasswing.
OpenAI’s new system, GPT-5.4-Cyber, will be accessible only to top-tier users and organizations participating in its Trusted Access for Cyber (TAC) program.
According to the company, the program includes “thousands of verified individual defenders and hundreds of teams responsible for defending critical software,” although no specific partners were named.
While not specifically designed for cybersecurity, Anthropic’s Mythos model impressed experts by identifying vulnerabilities in widely used software—some of which had remained undiscovered for years or even decades.
Reports also indicated that senior executives from major U.S. banks recently met with Treasury Secretary Scott Bessent and Federal Reserve Chairman Jerome Powell to discuss potential risks the technology could pose to the financial sector.
The release of such models comes amid increasing attention in Silicon Valley on generative AI’s growing ability to produce and analyze computer code. These capabilities allow systems to detect bugs and security flaws that could be exploited, even as developers attempt to implement safeguards to prevent misuse.
OpenAI said GPT-5.4-Cyber has been “trained to be cyber-permissive,” enabling defenders to test systems for vulnerabilities with fewer restrictions.
Anthropic has emphasized that limiting access to its Mythos model is intended to give defenders time to address vulnerabilities before they can be exploited.
“We don’t think it’s practical or appropriate to centrally decide who gets to defend themselves,” OpenAI said, adding that it aims to enable as many legitimate users as possible through systems that verify trusted use cases in an automated and objective way.
News.Az
By Nijat Babayev
