Researchers complain that OpenAI revoked their access to limited cyber program

Several security researchers said OpenAI revoked access to Trusted Access for Cyber (TAC), a limited program offering fewer restrictions on AI tools used for cybersecurity research. OpenAI confirmed the disruption resulted from an error. On August 19, 2026, researchers on OpenAI’s support forums and X said ChatGPT’s Cyber page reported that their identities could not be verified or their accounts were ineligible. TechCrunch spoke with five affected researchers. All said they lived outside the U.S. and Europe, suggesting a possible regional concentration, although the total number affected remains unclear. TAC provides vetted researchers with advanced models and fewer cybersecurity guardrails than those available to regular users. Applicants must submit identification and complete OpenAI’s verification process. The program is intended to help trusted defenders discover and report vulnerabilities for faster remediation, while limiting access for criminals seeking to develop exploits. One researcher shared an OpenAI email stating that Daybreak Blue access was revoked “due to a technical issue affecting a limited number of users.” OpenAI said the problem was on its end and asked affected researchers to reapply and complete verification. The company later said a limited set of users would need to re-verify to maintain access. Launched August 10, Daybreak Blue is the newest individual-researcher TAC tier. It provides frontier general-purpose models, including GPT‑5.6 Sol, with safeguards tailored to authorized defensive security work, including vulnerability discovery, secure code review, malware analysis, incident response and patch validation. OpenAI also introduced Daybreak Red, a higher tier for authorized vulnerability research, exploit validation and security testing. Anthropic operates a similar Cyber Verification Program (CVP). Researchers have criticized both companies’ guardrails as barriers to legitimate defensive and offensive security work. For investors, the incident underscores execution and trust risks in specialized AI programs, where reliable verification and misuse controls may influence professional cybersecurity adoption.