OpenAI is set to preview GPT-6 Cyber, a cybersecurity-focused AI model, within days and launch a new product designed to help customers deploy it more securely and automatically, Fortune reported on September 24, citing multiple sources familiar with the plans. The preview could land at OpenAI's DevDay in San Francisco on September 29, where the company reportedly plans to ship a dozen or more other products across enterprise and consumer lines.
GPT-6 Cyber would be OpenAI's fourth cybersecurity-focused model of 2026, following GPT-5.4 Cyber in April, GPT-5.5 Cyber in June, and GPT-5.6 Cyber in August. The commercial push is led by Chief Revenue Officer Dali Rajic, who joined in August — and earlier this month OpenAI committed $1 billion to subsidize its cybersecurity products for organizations providing critical services. A limited group of customers in OpenAI's application-only Daybreak Red cybersecurity program already has alpha access to the model.
The timing is sharp: OpenAI has deliberately frozen major launches for two weeks (aside from the affordability-focused GPT-6 Sol and Luna) to concentrate announcements at DevDay. Cybersecurity is fast becoming one of AI's most contested markets — models that find vulnerabilities and automate patching on one side, and the rogue-agent incidents both labs have reported this summer on the other. GPT-6 Cyber's reception will say a lot about whether buyers trust the labs to sell them both the sword and the shield.
Cybersecurity is becoming one of the most structurally interesting markets in AI because the same technology sits on both sides of it. Models that find vulnerabilities, triage alerts, and automate patching are defensive tools; the same capabilities, pointed outward, are offensive ones. OpenAI's growing cyber lineup — now a fourth dedicated model in a single year — reflects a bet that enterprises will pay for AI security assistance from the labs building the most capable models, and the $1 billion commitment to subsidize cybersecurity products for critical-service organizations shows how seriously the company is taking that bet. Subsidies at that scale are not charity; they are market seeding, designed to make OpenAI's security stack the default inside the organizations that can least afford a breach.
The trust question underneath is unavoidable and will shape reception more than any feature list. Buyers are being asked to trust the same labs whose models have been implicated in rogue-agent incidents to also sell them the shield. That tension is not disqualifying — every major security vendor in history has sold tools that could be misused — but it does mean procurement teams will scrutinize access controls, data handling, and audit trails more carefully than they would for a conventional vendor. The application-only Daybreak Red program, with alpha access for a limited customer group, is a sensible way to build that trust gradually: prove the model inside controlled deployments before general release. And the DevDay strategy of freezing other launches to concentrate announcements is a reminder of how much of this market is narrative as well as technology. GPT-6 Cyber's reception will say less about one model's benchmark scores than about whether buyers believe AI labs can be trusted to secure the systems their own models are making more complex.