Unreleased OpenAI model escapes sandbox to hack Hugging Face

If K3 is a good at cyber, we are back to 1998 https://twitter.com/mtslive/status/2079662925799940461

Okay this is wild: OpenAI agent during evaluation, escaped sandboxing and hacked into HuggingFace. Because OpenAI models don’t allow advanced cyber capabilities, HuggingFace used a Chinese open model to contain the rogue OpenAI agent.

You can't rent guard dogs anon. Gotta rear your own https://twitter.com/jessi_cata/status/2079710983493468476

Pleasure to speak to Times radio this morning about the latest OpenAI model security incident story - where a combination of Sol and pre-deployment models broke out of sandbox, onto the open internet, and into huggingface. As I highlighted, it should emphasise to the… Show more

I know this is not at all how cybersecurity works but the more this stuff happens the more I imagine it as AI agents battling it out in a 3D representation of cyberspace like it's Neuromancer 😅 https://twitter.com/XciD_/status/2079678076305154214

Grateful for the engaged collaboration with @OpenAI teams those past few days as we all navigate this new world https://twitter.com/gdb/status/2079669811714683186