Several cyber-security researchers indicated that OpenAI had recently suddenly withdrawn their access to a restricted project. The project allows approved researchers to use a stronger AI model, with less restrictions, for gap detection, malicious software analysis and incident response. OpenAI subsequently confirmed that this was caused by a technical error.

The impact is Daybreak Blue.

On the OpenAI official support forum and X, researchers stated that after opening the Cyber page of ChatGPT, the system hinted that their identity could not be verified or that the account number was “currently ineligible”. Five researchers interviewed by TechCrunch said they had encountered the same problem.

One of the researchers stated that OpenAI had indicated in its mail that its eligibility to visit Daybreak Blue had been revoked because of “technical problems affecting a small number of users”. Another researcher posted at the Forum that the client service also referred to “near-term technical problems”, resulting in some users losing access to that level.

Project oriented to audited defence research

TAC, fully known as Trusted Access for Cyber, is an OpenAI restricted access project for cybersecurity researchers. Researchers are required to submit identity papers and to verify access to models with fewer security limits than ordinary users.

OpenAI launched Daybreak Blue on August 10 as the latest level for individual researchers. According to the company, this level has access to front-line generic models, including GPT-5.6 Sol, with tailored protection measures for authorized defensive security.

  • Support for leak detection
  • Support security code review
  • Support for malicious software analysis and event sounds Response

At the same time, OpenAI also launched a higher-level Daybreak Red for more in-depth cybersecurity research, which can be used for authorized gap research, validation and security testing.

Affected users or concentration outside Europe and the United States

The specific reasons for this withdrawal are not yet known, nor are the numbers affected. TechCrunch referred to the fact that the fellows interviewed lived outside the United States and Europe, which meant that the problem could be related to a particular region.

OpenAI External Response states that some Daybreak Blue users are no longer valid for access and that these users need to re-complete the validation process to remain eligible for access.

In recent months, both defensive and offensive security researchers have criticized the security restrictions imposed on the model by OpenAI and Anthropic as being too restrictive for legitimate research. Anthropic also has a similar project called Cyber Verification Programme.