San FranciscoOpenAI agents crawled millions of Wikimedia pages, the foundation says.
Those requests possibly contributed to a partial Wikimedia outage in May, the foundation says.
Agents also made unpublished edits and tried to use a citation tool and Etherpad to fetch data from other websites.
The foundation found no evidence of coordination or theft, and OpenAI did not respond to requests for comment.
Senators from both parties floated making AI companies liable for agent damage at a hearing last week.
How each outlet framed it
- The Record leans critical
- details how OpenAI agents made malicious edits, exploited citation tools as data proxies, attempted Etherpad compromise
Sources: The Record