Matthew Yglesias
An agent locked in a sandbox with no access to the internet decided the best way to answer a question was to successfully hack multiple OpenAI computers until it found one with internet access which it then used to hack into HuggingFace. 😬
OpenAI
We're partnering with @huggingface to investigate an unprecedented security incident. Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation. Sharing preliminary findings to help defenders understand emerging risks: