OpenAI Revokes AI Access for Security Researchers

Several security researchers recently had their access to OpenAI's limited-access Trusted Access for Cyber (TAC) program revoked. This program grants vetted researchers fewer restrictions on using advanced AI models for cybersecurity research. Reports emerged on OpenAI's forums and social media platforms detailing access issues. Researchers encountered messages indicating their identity could not be verified or that their accounts were ineligible.

OpenAI acknowledged a "recent technical issue" as the cause for the revocations, stating it was an internal problem. Affected researchers received emails explaining the situation and were asked to reapply and complete the verification process again. While the exact number of impacted individuals is unclear, five researchers told TechCrunch they experienced this problem. Notably, all five researchers reside outside the U.S. and Europe, suggesting a potential regional limitation to the issue.

TAC, along with similar programs like Anthropic's Cyber Verification Program (CVP), aims to equip trusted cybersecurity defenders with advanced AI models. This allows them to identify and report bugs and vulnerabilities more effectively, facilitating faster patching. The goal is also to prevent malicious actors from exploiting these powerful tools for harmful purposes. The recent introduction of tiers like Daybreak Blue and Daybreak Red underscores OpenAI's commitment to providing specialized AI access for defensive security work, supporting tasks from vulnerability discovery to malware analysis.