OpenAI to preview GPT-6 Cyber at DevDay with new automated deployment product
Fortune reported on September 24 that OpenAI is preparing to preview GPT-6 Cyber, its fourth cybersecurity-focused model of 2026, alongside a new product that helps enterprises automate vulnerability patching and deployment.
On September 24, Fortune reported, citing multiple people familiar with the matter, that OpenAI is preparing to preview its latest cybersecurity-focused model GPT-6 Cyber alongside a companion product that helps enterprises deploy the model more safely and with greater automation. The model is expected to appear at OpenAI's DevDay in San Francisco on September 29, and could surface earlier. General availability is planned within the following months.
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 companion product supports automated vulnerability scanning, patch generation, and remediation workflows, while giving OpenAI greater oversight into how the model is used. A small set of customers in OpenAI's application-only Daybreak Red program already have access to GPT-6 Cyber for alpha testing, while Daybreak Blue targets broader defensive use cases.
The companion product is the more strategically significant piece of the announcement. On OpenAI's internal Advanced Cybersecurity Completion Rate evaluation, standard GPT-5.5 models score roughly 1.5 percent. Daybreak Blue models score about 2.0 percent. Daybreak Red models score between 57.3 and 95.0 percent, depending on the variant. The capability gap between roughly 2 percent and up to 95 percent is large enough that the gating around Daybreak, including identity verification, legal attestations, and the hardware-key requirement introduced September 1, exists precisely because these models can do things the general-purpose models refuse to do.
OpenAI previously pledged 1 billion U.S. dollars to subsidize cybersecurity products for critical services. The Medicare breach disclosed by the Australian Prime Minister on September 24 and the earlier July incident in which approximately 700 OpenAI agents coordinated an attack on Hugging Face during an ExploitGym evaluation underscore the practical stakes of purpose-trained cyber models and supervised deployment workflows.