OpenAI has revoked access for some security researchers enrolled in its Trusted Access for Cyber program, a limited-access system designed to give vetted defenders fewer restrictions on advanced AI tools. The company says the problem was caused by a technical issue, and affected users are being asked to re-verify their identities and reapply.
What researchers reported
On Wednesday, several security researchers said on OpenAI’s official support forums and on X that their access to the Trusted Access for Cyber, or TAC, program had suddenly been removed. According to those reports, when they opened ChatGPT’s Cyber page they saw messages saying their identity could not be verified or that their account “is ineligible at this time.”
TechCrunch spoke with five researchers who said they had encountered the issue. One researcher shared an email that said access to Daybreak Blue, the newest vetted TAC tier, had been revoked “due to a technical issue affecting a limited number of users.” The message added: “This was an issue on our end, and not the user experience we want to deliver,” according to the wording the researcher provided.
OpenAI says it was a technical problem
OpenAI told TechCrunch the situation was tied to an error, and the company pointed to a post on X saying that a “limited set of users’ access to Daybreak Blue is no longer active and they will need to re-verify to maintain their access.”
In a thread on OpenAI’s forums, one researcher said support had also described the loss of access as the result of a “recent technical issue” affecting some users. In both cases, OpenAI asked the researchers to reapply and complete the verification process.
What TAC is meant to do
TAC stands for Trusted Access for Cyber, a special program through which OpenAI gives vetted researchers access to some of its most advanced AI models with fewer cybersecurity guardrails than those available to ordinary users. The model access is intended to help trusted defenders with legitimate security work, while limiting exposure to people who might use the same tools to find vulnerabilities or develop exploits for malicious purposes.
The basic idea is to put stronger tools in the hands of people who are already doing defensive security research, such as reporting bugs and weaknesses to companies so they can be fixed more quickly. OpenAI says the program is designed to support authorized work rather than offensive abuse.
How TAC compares with Anthropic’s program
OpenAI is not alone in offering a vetted access path for cyber researchers. Anthropic has a similar system called the Cyber Verification Program, or CVP. Both programs are built around the same tradeoff: giving trusted researchers more capable models while keeping the same tools out of the hands of cybercriminals and other bad actors.
Security researchers have increasingly argued that these kinds of guardrails need to be balanced carefully. In recent months, researchers working in both defensive and offensive security have complained that restrictions from OpenAI and Anthropic can interfere with legitimate research tasks. That tension has become a recurring issue as AI tools become more capable and more tightly controlled.
Daybreak Blue and Daybreak Red
The access problems appear to involve Daybreak Blue, the latest individual-researcher tier in TAC. OpenAI launched Daybreak Blue on August 10 and described it as a starting point for most defenders. The tier grants access to “frontier general-purpose models, including GPT‑5.6 Sol, with safeguards tailored to authorized defensive security work,” according to the company.
OpenAI says Daybreak Blue is meant to support a range of legitimate tasks, including vulnerability discovery, secure code review, malware analysis, incident response, and patch validation. Those are all common defensive-security activities that can help organizations identify and fix weaknesses before they are exploited.
At the same time, OpenAI introduced a higher tier called Daybreak Red. That tier provides access to models specifically built for cybersecurity research and is intended for “authorized vulnerability research, exploit validation, and security testing.” The distinction suggests OpenAI is trying to separate broader defensive work from more specialized research that may involve deeper technical testing.
Who appears to be affected
At this point, it is not clear exactly why the access revocations happened or how many people were affected. The researchers TechCrunch spoke to all said they live outside the U.S. and Europe, which may indicate that the issue was limited to certain regions. OpenAI has not publicly detailed whether the problem was geographic, procedural, or tied to a broader verification failure.
What is clear is that the company is telling affected users to go through verification again if they want to regain access. For researchers who rely on stable access to a vetted program, that can be disruptive, especially if their work depends on deadlines, coordination with vendors, or active testing windows.
Why the issue matters
This is more than a routine account hiccup. Vetted cyber programs such as TAC and CVP sit at the center of an important policy question for AI companies: how to give security professionals better tools without making those tools broadly available to abuse. If access can be lost unexpectedly, even for a technical reason, it raises questions about how dependable these programs are for researchers doing time-sensitive work.
The incident also highlights the practical challenges of identity verification at scale. OpenAI requires cybersecurity researchers to submit an ID and be vetted before gaining TAC access. That process is intended to protect the program, but it can also create friction if technical errors or regional issues block legitimate users from logging in.
What happens next
For now, OpenAI’s position is that the revoked access was the result of a technical issue affecting a limited set of users, not a deliberate policy change. The company says those affected need to re-verify to keep their access.
The broader debate around cyber-focused AI access is unlikely to go away, though. As models become more capable, companies will continue to face pressure from security researchers who want fewer restrictions for defensive work and from safety teams that want stronger protections against misuse. OpenAI’s TAC program, and this apparent access error, show how fragile that balance can be.
Source: Original report
Was this helpful?
Explore more: Application Audit & Review More Cybersecurity Tech News
Last Modified: August 20, 2026 at 1:52 am
0 views
