OpenAI unveiled its new model, “GBT-5.4-Cyber,” which is a modified version of its latest leading model designed specifically for defensive cybersecurity purposes, following the announcement by Anthropic, the competing company that developed the “Cloud” system, about the advanced artificial intelligence model, “Mythos.” The new model will initially be offered on a limited scale to security service providers, organizations and researchers subject to audit, in a step that reflects a frenetic race among major companies in the field of employing artificial intelligence to serve cybersecurity.

Published on: 04/15/2026 - 12:04

OpenAI unveiled GPT-5.4Cyber, a modified version of its latest leading model designed specifically for defensive cybersecurity purposes, following competing company Anthropic's announcement of its advanced artificial intelligence model, "Mythos."

Anthropic announced Mythos on April 7 as part of the company's Project GlassWing, a controlled initiative under which selected organizations will be allowed to use the as-yet-unreleased Cloud Mythos Preview model for defensive cybersecurity purposes. The model discovered "thousands" of major vulnerabilities in operating systems, web browsers and other software.

Read also: "OpenAI" is preparing to ease restrictions on "GPT Chat" to allow pornographic conversations for adults

Details of the launch of the OpenAI model

OpenAI, the developer of the popular chatbot “ChatGPT”, said it would initially offer “GPT-5.4-Cyber” on a limited scale to vetted security service providers, organizations and researchers. The company added in a post on its website that it is also working to expand the “Trusted Access for Cyber” program to include thousands of individuals and hundreds of teams that protect vital software.

Graduated verification levels

OpenAI is adding new levels to its TAC program, which it launched in February, as higher levels of verification offer greater capabilities. Certified users at the highest level will have access to GPT-5.4-Cyber, which imposes fewer restrictions on sensitive cybersecurity tasks such as vulnerability research and analysis.

France 24/Reuters

Keywords related to the article

The content you want to browse is no longer in service or is currently unavailable.