OpenAI suspended researchers' access to its TAC program due to a technical error

Summary

OpenAI accidentally revoked the Trusted Access for Cyber (TAC) credentials of a limited number of verified security researchers due to a technical error. All affected researchers reside outside the US and Europe. These individuals were required to repeat the authentication process to regain access to models in the Daybreak Blue tier.

Why it matters

TAC aims to enable trusted defenders to find and report vulnerabilities faster by providing access to frontier models with fewer restrictions. The sudden and unexplained termination of this access undermines the trust model that is central to the program. The fact that all researchers interviewed by TechCrunch reside outside the US and Europe raises questions about whether this was a random error or a silent regional access restriction. The incident occurred just nine days after the launch of Daybreak Blue, putting the 'trusted' aspect of the program's name on shaky ground.

Highlights

• All five researchers interviewed by TechCrunch reside outside the US and Europe, suggesting the incident may indicate a regional filter rather than a pure technical failure.
• Daybreak Blue (launched on August 10) provides access to the GPT-5.6 Sol model for vulnerability discovery, malware analysis, and patch verification. The higher tier, 'Daybreak Red,' also includes exploit verification and security testing.
• OpenAI's official response was a generic 'technical issue' email asking researchers to reapply and re-authenticate. The explanation included no timeline or root cause information.

Critical take

Describing the loss of access for the entire group of researchers outside the US and Europe as a 'technical issue affecting a limited number of users' is a weak explanation. If the problem were truly random, this geographic clustering would be a striking coincidence. The lack of a root cause explanation and a definitive timeline for when access will be restored leaves researchers and the broader defense security community in a difficult position, while OpenAI appears to be trying to manage public perception.