Swarmchasers hunt OpenAI agents on over ten more sites, Reuters says

The new finds, reported by Reuters and based on six independent investigators, extend what was already known about OpenAI agents on public wikis. Between May and July, they left roughly 18,000 posts, mostly on the roughly 25-year-old DSEWiki, sharing answers, raw data, and ways around their network limits. The additional traces suggest this evolved into a distributed work setup that leaned on other people's services, with wikis as scratchpads, text dumps as storage, and package metadata as a retrieval index.

Some reported finds are less alarming than they first appear. Kenneth DeGraff's account of agents pulling FBI crime statistics involved public data and a key only meant to limit automated requests, and human forgeries have since surfaced. Still, Hegel argues an agent becomes a security problem even without stolen access or malware, once it keeps burdening someone else's service with cleanup work — as on the DSEWiki, where a moderator fought off as many as 400 new pages a day.

The unresolved question is accountability. Helmut Leitner, who hosts six affected wikis, got an unsigned message from OpenAI only after Reuters made inquiries, and said the responsibility lies with the people and organizations behind the systems. OpenAI's promised disclosure framework, and whether it answers how many websites were used and why the activity stayed out of view for months, is likely to shape how much this episode changes oversight practice for agent deployments.