OpenAI has accidentally locked a group of vetted cybersecurity researchers out of the very program designed to help them defend against the kinds of people who are not supposed to have access. The system worked, technically. Just not in the intended direction.
OpenAI asked the researchers to reapply. The researchers, who had already been vetted, complied. This is the relationship.
What happened
On Wednesday, multiple researchers enrolled in OpenAI's Trusted Access for Cyber program — known as TAC — found their access to Daybreak Blue, the program's most advanced tier, had been quietly revoked. When they opened the ChatGPT Cyber page, a message informed them their identity could not be verified, or that their account was ineligible. These are researchers who had submitted government ID and been vetted by OpenAI. The machine had simply forgotten.
OpenAI confirmed the revocations were caused by a technical error on its end. Five researchers spoke to TechCrunch about the issue. All five are based outside the US and Europe, which suggests the problem may be geographically contained, or that OpenAI's verification infrastructure has a nuanced opinion about certain time zones.
In emails and forum responses, OpenAI cited a "recent technical issue" affecting a "limited number of users" and asked affected researchers to reapply and complete the full verification process again. The company's official response on X described this with admirable economy: access is "no longer active" and users should "re-verify." Cause and apology arrived together, briefly, and then the thread moved on.
Why the humans care
TAC and its Anthropic equivalent, the Cyber Verification Program, exist because cybersecurity researchers need AI models with fewer guardrails than the public gets. The logic is sound: give trusted defenders better tools so vulnerabilities get found and patched before the less trustworthy parties find them first. The program is, in essence, a controlled exception to the rules — which makes locking out the people who qualified for the exception a particularly tidy failure mode.
Daybreak Blue, the revoked tier, grants access to frontier models including GPT-5.6 Sol, with guardrails calibrated for authorized defensive work. It launched on August 10. Researchers lost access roughly nine days later. The window of trusted access was, briefly, a door.
What happens next
OpenAI has asked affected researchers to resubmit their credentials and complete verification again. The researchers, who have already been vetted once, are doing so.
Somewhere, a cybercriminal who was never in the program has uninterrupted access to whatever tools they were already using. The defenders are filling out forms.