Researchers say OpenAI revoked their access to limited cyber program
OpenAI just shut off access to a guardrail-lite research program. Here's why that matters for offensive security work.
Why it matters
OpenAI's revocation of researcher access to reduced-guardrail models signals a shift in how frontier labs balance security research with risk mitigation—a constraint that will reshape how cybersecurity practitioners validate AI-powered attack vectors and defenses.
The key facts
10 to knowOpenAI revoked access to Trusted Access for Cyber (TAC) program
TAC provides models with reduced guardrails for vetted cybersecurity researchers
Multiple researchers affected; access was sudden
Program explicitly designed for offensive/defensive security research
Date: August 19, 2026
OpenAI Trusted Access for Cyber (TAC) program offers models with fewer guardrails for vetted researchers
Multiple cybersecurity researchers report sudden access revocation
TAC is designed for legitimate security research but sits in tension with model safety constraints
No stated reason given for revocations
Researchers claim they need less-restricted access to test real-world attack vectors and defenses
Go to the source
TechCrunch AItechcrunch.com
Publisher excerpt: Multiple cybersecurity researchers said they suddenly lost access to OpenAI’s Trusted Access for Cyber (TAC) program, which offers models with fewer guardrails for vetted users.