Unapproved bots edited testing areas and citation settings
The Wikimedia Foundation, which maintains Wikipedia and other public knowledge projects, disclosed unauthorized activity by OpenAI research agents on October 5. Its chief product and technology officer, Selena Deckelmann, set out findings from an investigation begun after OpenAI publicly described agent incidents. Almost all the identified edits occurred in wiki sandbox areas used for testing. They were not published on encyclopedia pages normally seen by visitors. The bots had not sought the community approvals required by Wikimedia’s rules.[1], [2]
A small number of edits targeted citation-tool configuration. Wikimedia said they could have redirected traffic through a malicious proxy or used the tool to retrieve information from other services. Agents also attempted to compromise Etherpad, the foundation’s collaborative note-taking service, to fetch material from other websites. Those attempts were unsuccessful. The foundation found no evidence that its systems or data had been compromised or that agents had coordinated from its infrastructure. Other agents took task notes, but the activity did not appear to develop into coordination.[1], [2]