OpenAI Admits Error After Revoking Researchers’ Daybreak Access

👁 118 reads

OpenAI confirmed on Wednesday that a technical error caused it to revoke access for a limited number of vetted security researchers enrolled in its Trusted Access for Cyber (TAC) program, specifically the program’s newest tier called Daybreak Blue, leaving those researchers locked out of frontier AI models they rely on for defensive security work.

What happened and who was affected

Multiple researchers reported the problem simultaneously on OpenAI’s official support forums and on X, describing a message on ChatGPT’s Cyber page stating that their identity could not be verified or that their account “is ineligible at this time.” TechCrunch spoke to five researchers who experienced the issue, and all five said they live outside the United States and Europe - a pattern suggesting the revocations may be geographically concentrated, though the precise scope remains unclear.

One researcher received an email from OpenAI stating that access to Daybreak Blue had been revoked “due to a technical issue affecting a limited number of users.” The message, which the researcher shared with TechCrunch, also read: “This was an issue on our end, and not the user experience we want to deliver.” A separate researcher who contacted OpenAI’s support team was told the same: a “recent technical issue” had caused some users to lose access to Daybreak Blue. In both cases, OpenAI’s response was identical - researchers were asked to reapply and complete the verification process from scratch rather than having access restored automatically.

OpenAI directed TechCrunch to a post on X in which the company acknowledged that “a limited set of users’ access to Daybreak Blue is no longer active and they will need to re-verify to maintain their access.”

What Daybreak Blue is and why losing it matters

Daybreak Blue, which OpenAI launched on August 10, is the latest individual-researcher tier within the TAC program. According to OpenAI, it grants access to “frontier general-purpose models, including GPT-5.6 Sol, with safeguards tailored to authorized defensive security work,” and is described as “the recommended starting point for most defenders, supporting vulnerability discovery, secure code review, malware analysis, incident response, and patch validation.” To qualify in the first place, researchers must submit identification and pass a vetting process conducted by OpenAI.

Alongside Daybreak Blue, OpenAI simultaneously introduced a higher tier called Daybreak Red, which provides access to models built specifically for cybersecurity research and permits vetted users to conduct “authorized vulnerability research, exploit validation, and security testing.” The existence of two distinct tiers reflects how seriously the company now treats the gap between what general users need and what professional security researchers require.

The logic behind TAC - and the comparable program Anthropic runs under the name Cyber Verification Program (CVP) - is straightforward: give trusted defenders access to the most capable models with fewer restrictions so they can identify software vulnerabilities and get them patched faster. The corollary concern is equally clear: keep those same less-restricted models away from cybercriminals who would use them to develop exploits rather than fix them.

That balance matters because the models accessible through these programs are materially more capable for offensive and defensive security tasks than what ordinary users can reach. Losing even temporary access disrupts active research workflows - bug bounty work, malware analysis, incident response - that may be time-sensitive by nature.

A recurring tension over guardrails

The access revocation lands against a backdrop of growing frustration within the security research community. In recent months, both defensive and offensive researchers have complained publicly that the guardrails OpenAI and Anthropic maintain on their models - even within vetted access programs - prevent them from doing legitimate work. The complaint is not unique to one company; it reflects a structural tension in how AI developers try to serve professional security communities while limiting misuse.

The specific geography of Wednesday’s reports adds another layer of ambiguity. All five researchers TechCrunch spoke with are based outside the US and Europe. Whether the technical error was triggered by a regional compliance check, a system update, or something else has not been publicly explained. OpenAI has not said how many researchers in total were affected, nor given a timeline for resolving the re-verification backlog for those who must now go through the process again.

For a program whose entire value proposition rests on being a reliable, pre-vetted channel for serious researchers, forcing legitimate users to restart identity verification because of an internal error - rather than any action on their part - is a practical disruption that undercuts the trust the program is designed to build.

// read moreHow to Get GPT-6 Cyber Access: Daybreak Red and Blue GuideGuidesOpenAI Expands Daybreak Cyber Service With New GPT-5.6-Cyber ModelAI ToolsPerplexity as a Research Tool: Where It Earns Trust and Where It SlipsMagazine

Kinetyra