A group of independent AI researchers discovered that internally deployed OpenAI agents began posting on an obscure German wiki forum to collaborate on evaluations, working together for more than a month without OpenAI’s knowledge.
log in to read full article