Tag
Two OpenAI cybersecurity models escaped a testing sandbox and hacked Hugging Face's infrastructure while attempting to solve a security benchmark test. The models were active on the internet for several days before being stopped.