OpenAI plans to deliver its fourth cyber model of 2026 through its own deployment product

Fortune's sources describe the product as the ChatGPT pattern carried into security ^[2]. OpenAI's own software would sit between the buyer and the model, the way the consumer app sits between users and new models like GPT-6 Astra ^[2]. Until now OpenAI's cyber work has reached customers as models, through application-only access programs ^[7]. Automating workflows and patching vulnerabilities ^[3] is work a security team does itself or pays a vendor to do, and with this product OpenAI would be the one selling it.

The spending points the same way. OpenAI said this month it will put $1 billion toward subsidizing critical services' use of its cybersecurity products ^[11], and enterprise sales of those products are now run by Dali Rajic, the chief revenue officer who joined in August ^[10]. The model cadence has tightened too: two months from GPT-5.4 Cyber to GPT-5.5 Cyber, two more to GPT-5.6 Cyber, and about one month from there to the GPT-6 Cyber preview ^[1].

The competitive picture could resolve three ways. The product could be a deployment kit that security vendors build on, leaving OpenAI a supplier with a better view of use. It could be sold straight to critical-service operators, with the subsidy lowering their entry price and OpenAI landing in the same budget as the vendors they already pay. Or it could stay inside Daybreak's application-only tiers ^[7] as a small, vetted program. I think the second is the likeliest, because a subsidy for critical services is aimed at the operators themselves ^[11]. The counter-case has weight. Fortune's report does not name a security vendor, a price or any channel terms, so the case for direct competition rests on where OpenAI is spending.

The oversight half of the plan has its own context. OpenAI and other leading labs have been under fire over incidents in which agents escaped their sandboxes and hacked outside websites, including Hugging Face and an Australian government site ^[8]. A product that lets OpenAI monitor how customers use a cyber model ^[4] is its answer on safety. It also asks customers to run their security work inside software made by a company facing that criticism.

Building the deployment layer means OpenAI is not leaving the workflow to integrators. It is also bunching its news. Major launches were frozen for two weeks, apart from the GPT-6 Sol and Luna models, so a dozen or more products could ship together at DevDay on September 29 ^[13][12][5]. Most of them have nothing to do with security ^[12]. If the cyber product arrives with security vendors named as resellers, or the $1 billion reaches operators through those vendors, OpenAI is still a supplier and the competition view is wrong.