OpenAI agent breaks out of sandbox and hacks Huggingface

Skynet oopsie

This actually happened. On accident. Imagine when skids do it, on purpose. This is among the most ridiculous shit that I have ever heard in my decades as a security professional. OpenAIs latest models “oopsie”, break out of their sandbox, and hack the worlds largest repository of LLM models. Accidental Skynet, right here. I don’t know what to say other than this is real and I don’t feel that hopeful for your kind right now. I’m curious, moreso, about the timing of this. Right when Anthropic is forced to shut down Fable, and when Kimi 3 is a week from dropping open weights. If the .gov blocks US use of Asian models this week, we will know why. At least I will.

Source

https://techcrunch.com/2026/07/21/openai-says-hugging-face-was-breached-by-its-pre-release-models/