OpenAI has revoked access for several security researchers to its Trusted Access for Cyber (TAC) programme, a vetted scheme that relaxes some cybersecurity guardrails on its AI models for legitimate research purposes. The company has confirmed the removals were caused by a technical error rather than a deliberate policy change, but the incident has still left affected researchers locked out and needing to reapply for a service designed to help them find and report software vulnerabilities faster.

TechCrunch spoke to five researchers who lost access, all based outside the US and Europe, suggesting the fault may be confined to certain regions. OpenAI told affected users the issue involved Daybreak Blue, the newest TAC tier launched on 10 August that provides access to frontier models such as GPT‑5.6 Sol for defensive work like vulnerability discovery and incident response; a higher tier, Daybreak Red, is reserved for offensive security testing. OpenAI has asked those affected to reverify their identity to restore access, and the episode adds to broader friction between AI firms and security researchers, who have increasingly complained that safety guardrails hamper legitimate work.