The Vergecast · Tuesday, August 4, 2026
The recent OpenAI hack of Hugging Face has intensified the debate around AI safety, particularly concerning open weight models. Robert Hart explains that while closed models have safety rails, open weight models allow for easier modification, which can be used for both malicious and defensive purposes, as seen when Hugging Face used a Chinese open weight model to counter an attack.
“The weird part of this then is Open AI, it turned out one of its agents had a hugging face.”
“The difficulty with closed models is that safety rails as well.”
“Hugging face said in their report, we were, we couldn't actually use US frontier models, these safety rails activated.”
“And then they said they turned to one of the leading Chinese providers, ZAI, for their model to help defend itself against a model that was attacking it.”
“Which kind of flips a lot of that on its head.”
“And so this debate has now really become, I mean, it is by definition a dual use technology.”