OpenAI "rogue" agent activities found on Wikimedia projects
Source
diff.wikimedia.org
Author
brokensegue
Date
Why it matters
Shows what unsanctioned AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → activity looks like from the defender's side: proxy abuse via a citation tool, probing of a public notepad and heavy crawling. Agent builders need to enforce disclosure and rate discipline.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
