OpenAI cuts researchers’ access to limited cyber program

Several security researchers report that OpenAI unexpectedly revoked their access to a limited program designed to relax certain AI usage restrictions for cybersecurity work. The company has acknowledged the disruption, attributing it to an internal error.

On Wednesday, a number of researchers took to OpenAI’s official support forums and X to document that their enrollment in the Trusted Access for Cyber (TAC) program had been terminated. Upon visiting ChatGPT’s Cyber page, these users encountered messages stating their identity could not be confirmed or that their account was “ineligible at this time.”

TAC serves as a specialized initiative where OpenAI grants vetted researchers entry to its most sophisticated AI models, but with considerably fewer cybersecurity guardrails compared to standard consumer access. Anthropic operates a comparable offering, the Cyber Verification Program (CVP) .

The core purpose behind TAC and CVP is to equip trusted security professionals with superior tools for identifying and reporting bugs and vulnerabilities to companies, thereby accelerating the patching process. Simultaneously, these programs aim to keep such powerful models out of the hands of cybercriminals who might otherwise exploit them to discover flaws and craft attacks. Prospective TAC participants must submit identification and undergo a vetting process by OpenAI.

The precise cause of these access revocations remains unclear, as does the total number of affected users.

TechCrunch interviewed five researchers who encountered this problem. One individual shared an email from OpenAI stating their access to Daybreak Blue, the latest vetted tier under TAC, had been revoked “due to a technical issue affecting a limited number of users.” The correspondence, provided to TechCrunch, added, “This was an issue on our end, and not the user experience we want to deliver.”

Similarly, a forum thread detailed a researcher who, after contacting official support, was told a “recent technical issue” had led to the loss of Daybreak Blue access. In both instances, OpenAI advised the researchers to reapply and undergo the verification process anew.

Notably, all five researchers who spoke with TechCrunch reside outside the U. S. and Europe, hinting that the problem could be geographically concentrated.

OpenAI directed TechCrunch to a company tweet acknowledging that a “limited set of users’ access to Daybreak Blue is no longer active and they will need to re-verify to maintain their access.”

Daybreak Blue, launched on August 10, represents the newest access tier for individual researchers, providing entry to “frontier general-purpose models, including GPT‑5.6 Sol, with safeguards tailored to authorized defensive security work.” OpenAI describes it as “the recommended starting point for most defenders, supporting vulnerability discovery, secure code review, malware analysis, incident response, and patch validation.” The company also introduced a higher tier, Daybreak Red, which offers models built specifically for cybersecurity research, enabling vetted users to conduct “authorized vulnerability research, exploit validation, and security testing.”

Lately, both defensive and offensive security researchers have voiced frustration over the guardrails implemented by Anthropic and OpenAI, contending that these restrictions hinder legitimate research efforts.