OpenAI Accidentally Revokes Cybersecurity Researchers' Access to Specialized AI Program

OpenAI unexpectedly pulled access for multiple cybersecurity professionals participating in its restricted security research initiative, attributing the widespread disruption to an internal systems glitch. The sudden revocations left vetted users locked out of specialized artificial intelligence tools designed for defensive security analysis.

The issue affected participants in OpenAI’s Trusted Access for Cyber (TAC) program, as first reported by TechCrunch AI. Affected researchers reported encountering error notices on ChatGPT's Cyber portal stating their identity could not be authenticated or that their accounts were currently ineligible for the service. Multiple users voiced their concerns across X and OpenAI’s official developer support forums.

TAC is designed to provide selected security experts with access to powerful frontier models featuring modified safety boundaries. By loosening standard guardrails for credentialed researchers, the initiative aims to accelerate vulnerability detection, malware analysis, and code auditing, enabling defenders to help organizations patch security flaws before malicious actors can exploit them.

The latest disruption specifically impacted access to Daybreak Blue, a tier rolled out by OpenAI on August 10. Designed as the primary entry tier for defensive security work, Daybreak Blue provides access to advanced systems such as GPT-5.6 Sol with specialized safeguards tailored for security work. OpenAI launched Daybreak Blue alongside Daybreak Red, an advanced tier aimed at deeper exploit validation and authorized security testing.

According to correspondence shared by affected researchers, OpenAI acknowledged that an operational error caused the access revocations. In an email sent to a participant, the company clarified that access was removed "due to a technical issue affecting a limited number of users," adding, "This was an issue on our end, and not the user experience we want to deliver."

To regain entry into the program, OpenAI instructed impacted security analysts to re-submit their verification credentials and complete the application process from scratch. OpenAI support personnel echoed similar guidance in public forum responses addressing the Daybreak Blue outage.

While the total number of impacted participants remains undisclosed, all five researchers interviewed by TechCrunch resided outside the United States and Europe. The geographical pattern suggests that the verification error may have been tied to specific regional server systems or international identity verification protocols.

The incident occurs amid ongoing friction between AI developers and the broader cybersecurity community. Over recent months, both offensive and defensive security researchers have noted that safety guardrails enforced by major AI labs—including OpenAI and Anthropic, which operates its own Cyber Verification Program—frequently impede legitimate security evaluations and vulnerability disclosure efforts.

OpenAI did not immediately respond to requests for comment regarding the duration of the outage or the timeline for processing re-submitted applications.