OpenAI has started to offer limited access to its new AI model, GPT-5.4-Cyber, which is designed to pinpoint software vulnerabilities. This rollout comes just a week after Anthropic launched its own competing model, Claude Mythos Preview.
Initially, this model is being made available to a select group within OpenAI's Trusted Access for Cyber program, reaching a few hundred users now, with plans to widen access to thousands soon.
GPT-5.4-Cyber aims to assist organizations in identifying and correcting security issues, and with fewer limitations in place, it allows for deeper system probing.
This step underscores the growing rivalry between OpenAI and Anthropic in the realm of advanced AI, particularly regarding cybersecurity.
As these models improve in terms of coding and vulnerability detection, worries are increasing about how they might be misused by cybercriminals and entities supported by state actors.
Anthropic's Mythos, which can independently find and exploit zero-day vulnerabilities across key systems, has already caught the attention of governments and financial institutions. U.S. officials, including those in the Treasury, have been emphasizing the need for organizations to take these types of tools seriously, indicating a rising urgency around AI-related cyber threats.
While both companies are marketing their models as protective tools, their swift advancements are transforming the global cybersecurity landscape.
For comments and feedback contact: editorial@rttnews.com
Business News
May 22, 2026 14:46 ET Minutes of the latest Fed policy session was the highlight of the week along with survey data on the U.S. housing market. In Europe, survey data signaled the trends in the euro area private sector. Further, consumer price inflation data from the U.K. was in focus. In Asia, various economic indicators from China drew attention to the health of the economy.