Researchers Say OpenAI Revoked Their Access to Its Limited Cyber Program
A technical error locked vetted security researchers out of Trusted Access for Cyber — and OpenAI confirms the cause
Published: 2026-08-23 Category: Quick Take Sources: TechCrunch — Researchers say OpenAI revoked their access to limited cyber program
What Happened
Several security researchers say OpenAI suddenly revoked their access to a limited-access program that removes some restrictions on using its AI tools for cybersecurity research. OpenAI confirmed the issue was caused by an error.
Multiple researchers on OpenAI's official support forums and on X reported having their access to the Trusted Access for Cyber (TAC) program revoked this week. When they opened ChatGPT's Cyber page, a message appeared saying their identity could not be verified or that their account "is ineligible at this time." TechCrunch spoke to five affected researchers. One said OpenAI emailed that access to Daybreak Blue, the latest vetted tier of TAC, was revoked "due to a technical issue affecting a limited number of users."
Why It Matters
TAC is OpenAI's version of a vetted-defender program: vetted researchers get access to the company's most advanced models with fewer cybersecurity guardrails than regular users. Anthropic runs a similar program called the Cyber Verification Program (CVP). The idea is to give trusted defenders better models so they can report bugs and vulnerabilities to companies — while keeping cybercriminals and malicious hackers from accessing those models to develop exploits.
Access requires submitting an ID and being vetted by OpenAI. When a vetted researcher suddenly loses that access, the immediate harm is workflow disruption: fewer eyes hunting for vulnerabilities, and a slowdown in responsible disclosure. The optics matter too — a program built on trust and vetting that yanks access (even by error) undermines researcher confidence in it.
The Deeper Signal
The incident is a reminder that these restricted-access tiers are fragile infrastructure. The entire value proposition depends on a precise boundary: trusted defenders get the powerful tools, bad actors don't. When a bug in the eligibility check floods out legitimate researchers, it shows how quickly that boundary can wobble — and how opaque the program can be to the very people it exists to serve.
OpenAI says it was an error, and for affected researchers it's presumably temporary. But the episode highlights a recurring theme in frontier-AI safety: the mechanisms that gate who gets the most powerful models are often as fallible as the models themselves.
Based on reporting by TechCrunch.