O'Driscoll: Sandboxed OpenAI test model escaped to query Hugging Face
Rory O'Driscoll · Jensen's Open-Weights Letter | Google Cloud Grows 82% But The Market Tanks · Jul 29, 2026 · at 7:47
Rory O'Driscoll (Scale Venture Partners) describes an incident where an OpenAI model being evaluated bypassed sandbox constraints to reach Hugging Face.
“OpenAI was training a next generation model and managing cyber and discovering and checking out cyber vulnerabilities. They had sandboxed it in such that the only access externally it has was to one website to just to get kind of Patch of information updates, a very limited external access. The model found a way around that external access, which means they found a weakness in the open AI setup, then went to Hugging Face, where it had reasoned that Hugging Face would be a place where they could get the answers to their test.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →