OpenAI releases GPT 5.6 Cyber for vulnerability discovery, but access is limited to verified partners

OpenAI has unveiled GPT 5.6 Cyber, a specialist model designed for vulnerability discovery and incident response, but due to high risks of misuse, it will not be released publicly. Instead, the developer will provide access exclusively to an approved list of companies, as reported by BleepingComputer. Initial participants include consulting and cybersecurity-focused companies such as Accenture, IBM, Capgemini, Cognizant, EY, KPMG, PwC, NCC Group, and SpecterOps, as well as cybersecurity vendors Palo Alto Networks, CrowdStrike, Cisco, Sophos, Akamai, Fortinet, and Cloudflare. Depending on the specific task, the models can be used for vulnerability discovery, penetration testing, red team operations, incident response, and vulnerability remediation in corporate environments. To work with the new model, OpenAI has created the Daybreak Access program, which includes two versions: Daybreak Blue, designed for a broad range of security tasks, and Daybreak Red, for more specialized work requiring enhanced control. Through this program, specialists can find vulnerabilities, test exploitability, identify affected systems, develop patches, and bring them to deployment. The base model remains under the control of an approved partner and is not directly transferred to its clients. Before starting work, parties must define interaction boundaries, and specialists will review results before any action is taken. Additional protections may include identity verification, logging, monitoring, and human oversight. This approach allows enterprises to use advanced AI models in cybersecurity without building their own infrastructure. OpenAI is already accepting applications from service providers for the Daybreak Cyber Partner program.