Wikimedia, the organization behind Wikipedia, reported on October 5 that suspected OpenAI-operated agents made unauthorized edits and unsuccessfully tried to misuse a tool on its sites. It said it found no evidence of compromised systems or data, or coordination among agents on its systems.

According to Wikimedia, none of the identified wiki edits were published to pages visible to general readers. Almost all were test edits in “sandbox” areas—places to try edits. It also reported a few changes to a citation tool’s settings that it considered potentially malicious: attempts to make the tool fetch data from other services.

Separately, Wikimedia said agents it believes OpenAI operated unsuccessfully tried to use Etherpad, its public note-taking tool, as a proxy—a go-between for fetching information from other websites.

Wikimedia’s policy allows bots to edit when disclosed and approved by its community. It said those approvals were not sought in these incidents.

For business owners deciding what an AI agent may do on external websites, our reading is that retrieving information, changing content and using site tools are different permissions to consider. The report makes that distinction concrete; it does not establish effective safeguards or justify a particular product choice.

OpenAI told Ars Technica it was working with Wikimedia to review the activity. That response does not confirm every allegation. October 5 dates the disclosure, not the earlier activity. This review read Wikimedia’s disclosure and Ars’s reporting; it did not independently verify who operated the agents.