Researchers say OpenAI revoked their access to limited cyber program — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI revoked access to its Trusted Access for Cyber (TAC) program for multiple vetted researchers, with OpenAI citing a "technical issue affecting a limited number of users" rather than disqualifying the accounts
- The affected tier, Daybreak Blue, launched on August 10 and grants vetted defenders access to "frontier general-purpose models, including GPT-5.6 Sol" with safeguards tailored for defensive security work such as vulnerability discovery and malware analysis
- OpenAI simultaneously introduced a higher tier called Daybreak Red that grants access to models built specifically for offensive cybersecurity research, including exploit validation and authorized security testing
- All five researchers who spoke to TechCrunch live outside the U.S. and Europe — a geographic pattern OpenAI has not publicly addressed despite framing the revocations as a system-wide technical error
- Affected researchers were told to reapply and complete the ID-based verification process again to regain access, though OpenAI has not disclosed how many users were impacted
- Anthropic runs a comparable program called the Cyber Verification Program (CVP), and both companies have drawn complaints from defensive and offensive researchers who argue guardrails on standard models block legitimate security work
Why it matters: OpenAI blamed a generic "technical issue" but every researcher who spoke to TechCrunch was based outside the U.S. and Europe — a geographic concentration the company hasn't explained. For these vetted defenders, losing access to models specifically built to accelerate vulnerability discovery and patching means real delays in defensive work at a time when researchers across both OpenAI and Anthropic's programs are already pushing back against guardrails they say obstruct legitimate security research.
Ask SkimNews
