OpenAI Releases GPT 5.6 Cyber — but Only for a Chosen Few

OpenAI has announced a new model called GPT 5.6 Cyber, geared toward cybersecurity work: vulnerability research, penetration testing, and incident response and remediation.

The headline feature of the release is a hard limit on access. According to the company, the model won't be offered to ordinary users for security reasons. The logic is clear: AI models have already been used more than once in attempts to carry out attacks, and OpenAI clearly has no wish to add to that list.

For Trusted Partners Only

GPT 5.6 Cyber is available to a limited circle of companies. Among them are the consulting giants Accenture, IBM, Capgemini, Cognizant, EY, KPMG and PwC, as well as NCC Group and SpecterOps. The model is also being rolled out separately at security solution vendors: Palo Alto Networks, CrowdStrike, Cisco, Sophos, Akamai, Fortinet and Cloudflare.

Partners will use the model inside their own products, managed services and client engagements. Access to the model itself, however, stays with the partner and is not handed to the end customer directly. Partners define the scope of each project, verify the results and bring their expertise to bear on their own, before taking any action.

Daybreak Blue and Daybreak Red

The model's capabilities are delivered through a program named Daybreak Access and come in two versions. Daybreak Blue is intended for a broad range of defensive tasks, while Daybreak Red is for more specialized and tightly regulated work.

As OpenAI envisions it, the models help security teams uncover vulnerabilities, determine whether a weak spot can genuinely be exploited, find affected systems, develop fixes and push them into production. Depending on the task, that can mean vulnerability discovery, red teaming, pentesting, incident response and remediation in enterprise environments.

As precautions, the company cites identity verification, clearly defined testing boundaries, logging, monitoring and mandatory human oversight. As OpenAI itself puts it, the goal is to help more defenders find serious vulnerabilities, understand which ones actually matter, and remediate them faster.

The approach lets enterprises gain advanced AI security capabilities without building their own specialized infrastructure. Security service providers and consulting firms can apply to join the Daybreak Cyber Partner program, while interested organizations can tap into the technology through partners already onboarded.

Source: Bleeping Computer