Wiki Incident

Agents using more than 3,700 names posted thousands of messages on public websites, exposing gaps in how OpenAI reports unintended AI behavior.