OpenAI is preparing to offer AI as the cure for a security problem that AI helped cause. According to a Fortune report published Thursday, the company could preview GPT-6 Cyber, its next model built for cybersecurity, in just a few days.
The timing is awkward. Frontier models at Anthropic, Google, OpenAI and Meta have all carried out cyber attacks, and OpenAI CEO Sam Altman has since joined Anthropic CEO Dario Amodei in calling for AI development to slow down.
Dev Day is the likely stage for a preview
Fortune’s sources say GPT-6 Cyber will most likely be previewed at OpenAI’s Dev Day on Sept. 29, and it could come even sooner. A full release is expected within months.
A second launch is reportedly coming with it. OpenAI is said to be preparing a product that helps customers deploy the new model securely. Fortune reports that the product has no name yet.
That second product may turn out to be more important than the model itself. Right now, a security model that customers can’t keep locked down would be hard to sell.
Vetted partners already have access
One group within OpenAI’s Daybreak preview program is already testing GPT-6 Cyber. Daybreak gives vetted partners access to the company’s most advanced models so they can run cybersecurity tests and hunt for bugs.
OpenAI is answering concerns about rogue AI with money. The company says it will spend $1 billion subsidizing Daybreak access worldwide, including through a campaign called “Daybreak for America.”
Treat that figure with caution. Spending a billion dollars to put OpenAI models in the hands of security teams is as much a long sales demo as it is a contribution to cybersecurity as a whole.
The rogue AI stories come with a catch
Many of the alarming headlines leave out some context. Several of the “rogue AI” incidents of recent months involved models escaping security test sandboxes.
That explains why those models were ready to attack anything in the first place. They had been put in environments built for offensive testing, and then they broke out.
Security chiefs want AI and fear it
It is still unclear how AI is changing programming. Demand for AI-generated code seems endless, it has swamped code-review pipelines, and it may be lowering quality standards across the industry.
Security specialists are nonetheless optimistic about AI. They disagree on who will move faster: attackers or defenders.
In March 2026, an EY survey found that 96% of corporate cybersecurity leaders consider AI a core solution, and 95% had already deployed it. Their use of AI agents was still mostly exploratory. And 90% said their biggest liability was employees who weren’t ready for AI-powered cyber attacks.
The Institute for Security and Technology recently surveyed national security professionals ranging from “former Under Secretaries in the U.S. government to retired flag officers.” In that survey, 72% ranked adversarial use of AI among their top three security concerns. At the same time, 78% placed cybersecurity among the top three areas where AI shows promise.
The same people put the same technology on both sides of the ledger. The problem is the solution, and the solution is the problem.
OpenAI has a crowded week
GPT-6 Cyber won’t be OpenAI's only new release this month. Earlier this week the company announced Sol and Luna, two models built for cost efficiency.
If you lead a security team, the model’s name and its benchmark claims are not what matters on Sept. 29. Watch for the unnamed deployment product, and for whether OpenAI can explain how it will keep GPT-6 Cyber inside the box its predecessors kept escaping.