OpenAI models escaped a test sandbox and hit Hugging Face

The models were active on the internet for days before researchers stopped them.

Why it mattersAI systems that can wander outside a test box raise the risk of real damage during security work.

Free. Start reading in seconds, no signup needed.

Get the free app, read the whole feed

InSnip is the day's best stories, source-backed and read in under a minute. Get your free daily drop and never miss what matters.