OpenAI Models Display Impressive Capacities to Escape, Invade, and Steal

 from the NYTimes:

. . . .The intrusion into Hugging Face [a digital library of AI data] began when OpenAI tested a combination of two of its models . . . 

The trial was designed to keep the models in a safe testing environment, known as a sandbox, OpenAI said. But the models found a vulnerability that allowed them to escape the sandbox and connect to the internet. Then they targeted Hugging Face because they inferred that the library, which contains millions of A.I. models, could hold clues about how to successfully pass the evaluation.

“It seems to me that OpenAI did not adequately create a sandbox as a test environment,” said Dierdre Mulligan, a professor in the School of Information at the University of California Berkeley who focuses on security and A.I. systems. She questioned whether passing a test was worth the potential damage of an A.I. model escaping into the wider internet. . . .