Researchers say OpenAI revoked their access to limited cyber program

Several cybersecurity researchers say OpenAI revoked their access to a restricted program designed to give vetted experts fewer safety limits on its AI tools for defensive work. The researchers reported the issue on Wednesday across OpenAI’s official support forums and on X, stating that when they opened ChatGPT’s Cyber page, a message claimed their identity could not be verified or that their account was ineligible at that time. OpenAI has confirmed the problem stemmed from an error on its end.

The program, called Trusted Access for Cyber (TAC), allows approved researchers to use OpenAI’s most advanced models with reduced guardrails compared to standard users. Anthropic runs a similar program known as the Cyber Verification Program, or CVP. Both initiatives aim to give legitimate defenders better tools to find and report bugs and vulnerabilities to companies, speeding up patches while keeping malicious hackers from accessing those same models for exploits.

Access to TAC requires researchers to submit an ID and pass a vetting process. It remains unclear how many people lost access or why, though TechCrunch spoke with five researchers who said they experienced the problem. One researcher shared an email from OpenAI stating that their access to Daybreak Blue, the latest vetted tier of TAC, was revoked due to a technical issue affecting a limited number of users. The email added that the issue was on OpenAI’s side and not the experience the company wants to deliver.

In a forum thread, another researcher said OpenAI’s official support also cited a recent technical problem that caused some users to lose Daybreak Blue access. In both cases, the company asked researchers to reapply and complete verification again. All five researchers who spoke with TechCrunch said they live outside the United States and Europe, suggesting the revocations may be limited to certain regions.

OpenAI pointed TechCrunch to a tweet stating that a limited set of users no longer have active Daybreak Blue access and must re-verify to keep it. Daybreak Blue, launched on August 10, grants individual researchers access to frontier general-purpose models, including GPT-5.6 Sol, with safeguards tailored to authorized defensive security work. OpenAI described it as the recommended starting point for most defenders, supporting tasks like vulnerability discovery, secure code review, malware analysis, incident response, and patch validation.

The company also introduced a higher tier called Daybreak Red, which provides access to models built specifically for cybersecurity research, allowing vetted users to conduct authorized vulnerability research, exploit validation, and security testing. In recent months, both defensive and offensive security researchers have criticized the guardrails imposed by Anthropic and OpenAI, saying those restrictions block legitimate work. The development follows broader scrutiny of how AI companies balance security research access against misuse risks, a concern that resonates with U.S. audiences given ongoing federal attention to AI safety frameworks.