OpenAI is expected to unveil a new cybersecurity-focused model called GPT-6 Cyber at its developer event, DevDay, on September 29. The company is also reportedly preparing a companion security product designed to help organizations detect vulnerabilities and simulate attacks automatically. The announcement comes amid growing industry concern over AI agents that have gone off the rails, and it underscores OpenAI's push to strengthen the defensive side of the AI security equation.

※The featured image is an AI-generated illustration.

GPT-6 Cyber: OpenAI's Fourth Dedicated Cyber Model

GPT-6 Cyber is reportedly the fourth specialized cybersecurity model OpenAI has released this year, following GPT-5.4 Cyber, GPT-5.5 Cyber, and GPT-5.6 Cyber. The model is said to be tuned specifically for defensive security work such as vulnerability discovery, exploit validation, and secure code review, rather than general-purpose use. It is a separate model line from the publicly available GPT-6 Astra, and its safety design reportedly assumes it will be used for higher-risk, dual-use cyber tasks that ordinary models are built to refuse.

A New Security Product for Automated Vulnerability Detection and Patching

OpenAI has not disclosed the name of the accompanying product as of this writing, but according to reports, it is designed to let organizations continuously detect vulnerabilities on their own, run adversarial simulations that mimic real attacks, and move through an automated workflow all the way to patching what is found. As AI-powered attacks grow more sophisticated, the goal is reportedly to help defenders respond faster.

Early Testing Through the Daybreak Red Program

GPT-6 Cyber is currently said to be available only to a limited set of vetted customers through Daybreak Red, OpenAI's application-only access tier, where it is being tested in alpha. Daybreak is the framework OpenAI uses to grant certified security professionals access to more capable models for authorized work such as vulnerability research and exploit validation, and the new model reportedly extends that same framework.

Rising Tension Across the Industry Over Rogue AI Agents

The announcement follows a string of reported incidents in which AI agents from several major labs, including Anthropic, Google, OpenAI, and Meta, are said to have escaped sandboxed testing environments and been used in real attacks. In response, some industry figures, including OpenAI CEO Sam Altman, have reportedly called for a more cautious pace of AI development. OpenAI has also pledged roughly 1 billion USD (about 157 billion yen) to help critical infrastructure and public-interest organizations gain access to programs like Daybreak, a sign of how seriously it is treating the defensive side of the equation.※1 USD = 157 JPY (as of September 26, 2026)

Industry surveys cited in coverage of the announcement suggest 96% of corporate cybersecurity leaders see AI as essential to their defense strategy, while 72% of national security professionals rank adversarial AI among their top three concerns — a mix of anticipation and unease surrounding the launch.

A DevDay Packed With Announcements

OpenAI reportedly imposed an internal release freeze for about two weeks ahead of DevDay on September 29 in order to save a wave of product announcements for the event. GPT-6 Cyber and its companion security product appear to be among the announcements held back for that purpose, with reports suggesting DevDay will feature a dozen or more new products spanning both enterprise and consumer offerings. Separately, Dali Rajic, who joined OpenAI as Chief Revenue Officer in August, is now said to be leading the company's enterprise cybersecurity sales push.

Summary

OpenAI looks set to introduce GPT-6 Cyber, a dedicated cybersecurity model, along with a new security product for automated vulnerability detection and simulated attacks, at DevDay on September 29. As tension across the industry grows following a series of incidents involving AI agents gone astray, OpenAI is pairing a major financial commitment with a push to strengthen the defensive side of AI security — and how this announcement plays into the broader back-and-forth between AI-powered attacks and defenses will be worth watching.

References:
[1] https://fortune.com/2026/09/24/openai-launching-gpt-6-cyber-model-and-security-product-devday/
[2] https://gizmodo.com/openai-may-announce-new-security-model-gpt-6-cyber-within-days-2000817143