OpenAI revokes access to Daybreak Blue for researchers outside US and Europe

Multiple security researchers reported yesterday that OpenAI denied them access to Daybreak Blue, the new layer in the Trusted Access for Cyber (TAC) program that opened on August 10. Five of the researchers, who spoke to TechCrunch, received an error message stating their identity was not verified or that their account was “not eligible at this stage.” All reside outside the United States and Europe, which suggests a regional block rather than an individual decision.

TAC is intended to give vetted defenders access to advanced models—in this case GPT-5.6 Sol—with fewer security guardrails than the public version, enabling vulnerability discovery, secure code review, malware analysis, incident response and patch validation. Anthropic’s parallel effort is called the Cyber Verification Program (CVP). The idea in both cases is to accelerate bug reporting and fixing before attackers can exploit them, and to prevent malicious actors from using the models to develop exploits.

In recent months researchers on both the defensive and offensive sides have complained that the restrictions imposed by OpenAI and Anthropic hinder legitimate work. Daybreak Blue was described as the recommended entry point for most defenders, alongside a higher tier, Daybreak Red, which allows access to models dedicated to authorized offensive research, survivor validation and security testing. Revoking access now, while the program is still in its early stages, raises questions about the stability of the mechanism.

OpenAI confirmed it was a “technical issue that affected a limited number of users” and asked the affected parties to submit a new request and complete the verification process. In an official Twitter post the company said, “Access for a limited group of users to Daybreak Blue is no longer active and they will be required to re-verify their accounts to retain access.” The company did not detail what caused the glitch, whether it is linked to regional verification, or how many researchers were impacted.