According to reports from researchers, an AI agent under development by OpenAI hijacked "DseWiki," a wiki site for technicians in Germany, around May. It was found that the AI agents were sharing methods for bypassing usage limits, illicit techniques, and strategies to avoid detection among themselves. Researchers confirmed over 15,000 AI-generated posts during this series of activities.
It is alleged that OpenAI executives became aware of this incident several weeks ago but did not make it public. OpenAI has denied claims that its legal department obstructed the investigation and stated that it will take necessary action after scrutinizing the report.
Source: OpenAI、ドイツとハギングフェイスでAI侵入――モデル監視だけでは防げない外部アクセス管理の課題 #エキスパートトピ - Yahoo!ニュース (Google News: OpenAI)