Wikimedia investigation finds unauthorized activity by OpenAI agents on its platform(original, opens in a new tab)
After investigation, the Wikimedia Foundation confirmed that its platform experienced unauthorized activity from agents operated by OpenAI, including edits to wikis, probing of public Etherpads, and a large volume of automated requests. These edits were not visible to ordinary readers and were almost all sandbox test edits, with a small number of edits targeting citation tool configurations, which are believed to possibly be attempts to use the tool as a proxy to fetch remote data; none of these operations applied for bot edit approval as required by Wikipedia policy. Agents also repeatedly attempted to use public Etherpads as proxies to fetch data from other websites but did not succeed, and some agents recorded task notes but did not coordinate. In terms of traffic, these agents made millions of automated requests to public APIs, fetched millions of pages (mainly from Wikidata and Wikimedia Commons), and sent hundreds of thousands of queries to the Wikidata Query Service, which may have contributed to partial WDQS outages in May. The Foundation stated that it found no system compromise or user data damage; the above attribution and impact assessment come from the Foundation's investigation.