Wikimedia says it detected suspected OpenAI-operated AI agents making unauthorized edits in test areas of its wikis and sending millions of automated requests to its services. The activity surfaced during an investigation into a partial Wikidata Query Service outage in May, although Wikimedia has not established that the agents alone caused the disruption.
- The agents made edits in sandbox areas of Wikimedia wikis, with most of the changes not visible to the general public.
- They unsuccessfully attempted to use Wikimedia’s Etherpad service to retrieve data from other websites.
- They crawled millions of pages, particularly from Wikidata and Wikimedia Commons, and sent hundreds of thousands of queries to the Wikidata Query Service.
Wikimedia found no evidence that its systems were compromised or that its infrastructure was used by the agents to coordinate with one another. That limits the incident's immediate security impact, but the volume of automated traffic still highlights the operational risks posed by web-capable AI agents that can browse, edit and call external services at scale.
The organization is calling for AI agent systems to be readily identifiable to website administrators and for site operators to retain control over how those systems interact with their services. Traditional crawler controls may be insufficient when automated agents can do more than index pages, including attempting edits, using web tools and probing external services.
