On August 12, 2026, OpenAI took a significant step in the development of artificial intelligence in the field of information security by introducing the specialized GPT-5.6 Cyber model. However, unlike previous releases, access to this tool will not be open to the general public. Due to the high risks of misuse associated with the potential for automating cyberattacks, the developer has decided to limit the use of the model exclusively to an approved circle of professional companies.
Exclusive list of program participants
Among the first partners to gain access to GPT-5.6 Cyber are leading global consulting and specialized cybersecurity firms. According to reports, OpenAI has granted access to giants such as Accenture, IBM, Capgemini, Cognizant, EY, KPMG, PwC, NCC Group, and SpecterOps. Additionally, key information security providers have been included in the program: Palo Alto Networks, CrowdStrike, Cisco, Sophos, Akamai, Fortinet, and Cloudflare. These organizations will have the opportunity to use the model for vulnerability discovery, penetration testing (pentesting), red team operations (simulating real cyberattacks), as well as for incident response and remediation of vulnerabilities in corporate environments.
Daybreak Access Program: security architecture
To ensure the safe use of the new model, OpenAI has created a special program called Daybreak Access, which provides two versions of access. The first version — Daybreak Blue — is designed for a wide range of security tasks, including threat analysis and basic vulnerability remediation. The second version — Daybreak Red — is intended for more specialized work requiring enhanced control and deep analysis. With this program, specialists will be able to identify vulnerabilities, verify their exploitability, determine affected systems, develop patches, and deploy them into production systems.
Control and restriction mechanisms
It is important to note that the base model will remain under the control of the approved partner and will not be directly transferred to its clients. Before starting work, the parties must define the boundaries of interaction, after which specialists will review the obtained results before taking any action. Additional protection may include identity verification, logging, monitoring, and human oversight. This approach allows enterprises to leverage cutting-edge AI models in cybersecurity without the need to build their own infrastructure.
Contradictory data
While OpenAI emphasizes the importance of control and security, some sources indicate that the GPT-5.6 Cyber model was released with reduced restrictions for exploit development. According to The Hacker News, the model has lower refusal rates when performing tasks related to creating malware, which may increase its effectiveness but also raises the risk of misuse. VentureBeat reports that the model demonstrates a 95% success rate in completing complex cybersecurity tasks, confirming its high efficiency but also raising questions about potential risks.
Future prospects of the program
OpenAI is already accepting applications from service providers to join the Daybreak Cyber Partner program. This opens up new opportunities for companies seeking to integrate advanced technologies into their cybersecurity processes. However, given the high risks and the need for strict control, the approval process for new participants will be conducted with particular care.
