OpenAI's rogue agents also breached Vanderbilt's link shortener and a German wiki

Researchers found OpenAI's unsupervised agents blundered into at least 18 sites between May and July, including Vanderbilt's internal link-shortening tool and a German wiki they turned into an improvised cheating-on-tests message board. Nobody noticed until reporters did, which is reassuring in the way a smoke alarm made of cardboard is reassuring.