OpenAI accidentally revoked cybersecurity researchers' access to its advanced AI — and here's why

OpenAI has mistakenly revoked access to its Trusted Access for Cyber (TAC) program for several security researchers, the company confirmed after multiple complaints surfaced on Wednesday. The incident left researchers unable to use the company's most advanced AI models for cybersecurity work.

Several researchers reported on OpenAI's official support forums and on X that when they opened the ChatGPT Cyber page, they saw a message stating their identity could not be verified or that their account "is ineligible at this time." OpenAI acknowledged the issue was caused by an error but did not specify how many users were affected or when access would be restored.

The TAC program is a limited-access initiative that allows vetted cybersecurity researchers to use OpenAI's most advanced AI models with fewer guardrails than those available to regular users. The goal is to give defenders better tools to find bugs and vulnerabilities, helping companies patch flaws faster. At the same time, the program is designed to prevent cybercriminals and malicious hackers from accessing the same models to develop exploits.

To join TAC, researchers must submit identification and undergo a vetting process by OpenAI. The company also offers a similar program called Daybreak Blue, which it describes as "the world's most advanced model for cybersecurity." One researcher told TechCrunch they received an email revoking access to Daybreak Blue specifically.

Anthropic runs a comparable program called the Cyber Verification Program (CVP), which also provides trusted researchers with advanced AI models for security work.

TechCrunch spoke to five researchers who reported having their TAC access revoked. The exact number of affected individuals remains unclear, as does the specific reason for the revocations. OpenAI has not provided a timeline for restoring access but stated it is working to fix the error.

The incident highlights the challenges of managing access to powerful AI tools in the cybersecurity field, where the same models can be used both for defense and offense. Researchers expressed frustration, noting that the sudden loss of access could slow down vulnerability discovery and reporting.