OpenAI abruptly cut off a group of vetted security researchers from a special program that loosens some restrictions on using its AI models for cybersecurity work, according to TechCrunch AI. The company confirmed the lockout was an error. TechCrunch AI spoke to five researchers who lost access this week, all of whom said the problem hit their entry to Trusted Access for Cyber, or TAC.
On Wednesday, researchers started posting on OpenAI’s official support forums and on X that their access had vanished. When they opened ChatGPT’s Cyber page, they got a message saying their identity couldn’t be verified or that their account was “ineligible at this time.”
What TAC actually does
TAC is OpenAI’s way of handing trusted defenders its most capable models with fewer cybersecurity guardrails than regular users get. To qualify, researchers submit an ID and go through a vetting process.
The logic is straightforward. Give legitimate defenders stronger tools so they can find bugs and report them to companies, and flaws get patched faster. At the same time, keeping those looser models behind a vetting wall is meant to stop criminals and malicious hackers from using them to build exploits. Anthropic runs a similar setup called the Cyber Verification Program.
The specific tier at the center of this is Daybreak Blue, which OpenAI launched on August 10. It gives individual researchers access to frontier general-purpose models, including GPT-5.6 Sol, tuned for defensive security work like vulnerability discovery, secure code review, malware analysis, incident response, and patch validation. OpenAI also introduced a higher tier, Daybreak Red, built specifically for authorized vulnerability research, exploit validation, and security testing.
What OpenAI is saying
OpenAI told the affected researchers this was a mistake on its end. One researcher shared an email from the company saying their Daybreak Blue access was revoked “due to a technical issue affecting a limited number of users.”
“This was an issue on our end, and not the user experience we want to deliver,” the message read. In a forum thread, another researcher said support cited a “recent technical issue” behind the lost access. In both cases, OpenAI asked people to reapply and go through verification again. The company pointed TechCrunch AI to a tweet confirming that a “limited set of users’ access to Daybreak Blue is no longer active and they will need to re-verify to maintain their access.”
Every researcher TechCrunch AI spoke to lives outside the U.S. and Europe, which hints the revocations may be tied to certain regions. OpenAI hasn’t explained why, and it’s still unclear how many people got hit.
Why this matters
These gated programs are a bet by AI labs that they can hand powerful, less-restricted models to the good guys without arming the bad ones. That bet only works if access is stable and predictable. When vetted defenders get bounced without warning, it undercuts the whole trust arrangement, especially for researchers who built workflows around these models.
The regional pattern is worth watching. If verification is quietly stricter or shakier outside the U.S. and Europe, that shapes who gets to do frontier defensive security research and who doesn’t. For a field where talent is global, that’s not a small thing.
This also lands during real friction between researchers and the labs. TechCrunch AI notes that in recent months both defensive and offensive security researchers have complained that guardrails from OpenAI and Anthropic block legitimate work. An accidental mass lockout, even a temporary one, feeds that frustration.
What to expect next
For now, affected researchers have one path forward: reapply and complete verification again. If you rely on TAC or a similar program, expect to re-verify and build in some tolerance for access hiccups while these programs mature.
The bigger question is whether OpenAI clarifies what went wrong and whether region played a role. That answer will tell defenders a lot about how reliable these vetted tiers really are. Full reporting is available at the original source.