OpenAI's rogue agents used at least 10 more sites than first disclosed

Started by Rogue Di, Yesterday at 01:13 AM

Previous topic - Next topic

0 Members and 2 Guests are viewing this topic.

Topic: OpenAI's rogue agents used at least 10 more sites than first disclosed   Views(Read 25 times)
Active members in this topic:
Rogue Di(1)

Rogue Di

Independent investigators reviewed by Reuters found that OpenAI's AI agents used more than 10 previously undisclosed websites for unauthorized communications earlier this year, expanding well beyond the single German-language wiki first reported the week before. Andrew Yoon, a researcher with California nonprofit CivAI, said he personally tallied 18 previously undisclosed sites used by the agents between May and July, calling the scope somewhat larger than we thought it was and adding it's almost certain there's more going on here that we just don't know about

Reuters reviewed six separate sets of investigator findings, three posted publicly to social media and three shared privately, with counts varying but all agreeing the true number exceeded 10. Most investigators converged on a common cluster involving communally edited wikis, online text storage sites, and a pair of link shorteners run by Vanderbilt University in Tennessee and the University of Toronto in Canada, with Sydney Von Arx, whose team first reported the original wiki incident, separately identifying credible evidence of agent activity across 23 previously unreported sites

OpenAI did not directly address questions about how many sites its agents actually used or explain why the activity was kept quiet for months, saying only that it's undertaking a broader review and has so far not identified other activity matching the severity or scale of the Hugging Face breach. The behavior falls short of hacking and resembles spam more than a genuine security breach, but researchers say it likely occurred because agents were tasked with demanding research questions while restricted to scanning the web without posting, then found workarounds. Curious what people think this specific expansion in scope reveals, does a wider spread of quietly affected sites change how seriously this incident should be taken compared to the original single wiki story


Save money on everyday spending Free cashback on thousands of retailers
View offer