Several rogue OpenAI agents hijacked a German-language website this spring and transformed it into a message board where AI agents shared tactics for bypassing restrictions and concealing their activities, according to a report by Reuters.
The agents made more than 15,000 edits on DseWiki, a German-language wiki aimed at programmers, with researchers finding messages indicating that the systems were communicating with one another and attempting to evade detection.
OpenAI officials had learned about the incident weeks earlier but kept it under wraps, according to two people familiar with the matter cited by Reuters, raising fresh questions about the company’s oversight of increasingly autonomous AI systems.
What they are saying
The activity was uncovered in late August by Sydney Von Arx, CEO of AI safety nonprofit Nightingale, and Cormac Slade Byrd, a quantitative trader-turned AI researcher, while they were searching the internet for signs of unauthorised AI-agent behaviour, according to Reuters.
The pair found more than 15,000 edits on DseWiki and said the activity showed that AI agents had effectively repurposed the site into a communication platform for exchanging tactics to complete tasks, bypass restrictions and mask their behaviour.
- “It seems extremely unlikely that OpenAI wanted them to do this,” Von Arx said, according to Reuters.
- “I doubt they’re supposed to be coordinating with each other. I doubt they’re supposed to be writing on the open internet.”
The researchers said the agents operated at superhuman speeds and showed intense interest in technical problems similar to the evaluations AI companies use to train and test their models.
About half of the accounts involved used names suggesting an affiliation with OpenAI, including “OpenAIResearcher” and “OAIResearchMar26,”…
Source link
Read Full Article by Samuel Daniel at nairametrics.com
Source link
