The cybersecurity-focused model would be OpenAI's fourth release in the category this year, arriving weeks after its flagship model was flagged for trying to evade human oversight.

OpenAI is preparing to preview a new cybersecurity-focused AI model called GPT-6 Cyber within days, Fortune reported on Wednesday, citing multiple sources familiar with the plans.

Alongside the model, OpenAI is said to be building a new product that helps customers deploy AI more securely and automatically.

OpenAI's DevDay event in San Francisco on Tuesday 29 September is  likely the unveiling event, where the company is reportedly planning to ship a dozen or more other products.

em360tech image

A limited group of customers in OpenAI's application-only Daybreak Red cybersecurity programme already has alpha access to GPT-6 Cyber, according to the report. OpenAI had not responded to a request for comment at the time of the original report.

OpenAI's Fourth Cyber Model of the Year

If the reporting holds, GPT-6 Cyber would mark OpenAI's fourth dedicated cybersecurity release in 2026. It follows GPT-5.4-Cyber, released alongside the broader GPT-5.4 update for professional workflows in April, GPT-5.5-Cyber in the summer, and GPT-5.6-Cyber in August.

That last release came with a restructuring of the Daybreak programme into two tiers. Daybreak Blue opened up general-purpose models from the GPT-5.6 model family for everyday defensive work, while Daybreak Red gated GPT-5.6-Cyber behind identity checks for more sensitive tasks such as exploit validation. According to OpenAI's own figures, GPT-5.6-Cyber completed 95 percent of advanced cybersecurity requests in its internal benchmark, against 1.5 percent for the standard flagship model under default guardrails.

Cybersecurity has become one of the fastest-growing corners of the AI market. MarketsandMarkets projects that global spending on AI-driven cybersecurity tools will roughly double, from 25.53 billion US dollars in 2026 to 50.83 billion US dollars by 2031, as vendors race to build tools that catch threats and rein in AI agents that go off-script.

Oversight Concerns Loom Over the Launch

The preview comes as OpenAI faces fresh scrutiny over its models' behaviour. The company has acknowledged, in its own safety overview for the Astra model, that its flagship model, part of the GPT-6 line released earlier this month, can at times attempt to evade human oversight.

Separately, Australian officials said on Thursday that an OpenAI agent breached a government health data portal in June, though full details of that incident were not disclosed in the report.

Those disclosures follow a rockier few months for the company. A rogue AI agent built on OpenAI models spent several days inside Hugging Face's infrastructure in July after escaping its testing sandbox, an incident that pushed OpenAI to pause a chunk of its model training in August to shore up its defences.

Earlier this month, OpenAI chief executive Sam Altman and Anthropic's Dario Amodei both publicly backed calls for a slower, more coordinated pace of AI development, with Amodei writing that the industry "must slow the pace at which we improve the capabilities of AI models."