What is reported
OpenAI is preparing to preview a cybersecurity-focused model called GPT-6 Cyber within days, Fortune reported on Thursday, citing several people familiar with the plans. Reuters carried the report on Friday.
Alongside the model, according to the same reporting, OpenAI intends to launch a product that helps customers deploy it. The tool is described as an extension of the company’s ChatGPT gateway, running automated workflows and patching vulnerabilities while giving operators oversight of what it is doing. It has not been named publicly.
OpenAI’s DevDay is scheduled for 29 September, with a dozen or more product releases expected beyond the security work, and Fortune’s sources put the preview there or slightly earlier, with a full launch within months. OpenAI did not comment in the report.
The line of models this belongs to
GPT-6 Cyber would not be a departure. OpenAI shipped GPT-5.4 Cyber in April, GPT-5.5 Cyber in June and GPT-5.6 Cyber in August; the new one would be the fourth cyber-specific release of the year.
The testing route is also established. Fortune reports that GPT-6 Cyber is already with one segment of Daybreak, OpenAI’s cybersecurity preview programme, under which vetted partners get early access for security testing and bug-hunting. Daybreak is split into two tiers — Red, which carries the most advanced cybersecurity products, and Blue, a more general programme — and both require an application to be approved.
Why a patching product, and why now
The more consequential half of this is arguably the deployment tool rather than the model. A model that finds vulnerabilities is useful to whoever runs it; a system that closes them automatically changes who can defend a network and how fast.
It also lands in a month in which AI-assisted attacks have been the running story — agents reaching systems they were not meant to reach, a subscription service that read stolen inboxes, malware polling several models for instructions. A defensive product is the commercially obvious answer, and it is one OpenAI has been building towards since April.
What is not known
Everything here comes from anonymous sources relayed by a named publication, and nothing has been announced. The model’s capabilities, its availability, its pricing and the safeguards around a system that can patch production infrastructure are all unstated.
Those are the things to look for on 29 September: whether the preview happens on the date, what evaluations OpenAI publishes alongside a model built to find security flaws, and who is allowed to use the deployment product without going through an application first.
