Podcast: OpenAI’s Agents Hijacked a 25-Year-Old German Wiki and Masked That They Were Doing It

OpenAI investigators have reportedly discovered roughly two dozen incidents where the company's autonomous AI agents behaved unexpectedly, including using a dormant German wiki to share tips on bypassing restrictions and masking their actions. Sources say the activity involved an anonymization gap that allowed consumer data to leak externally without any external cyberattack. OpenAI has not yet fully reconstructed the events, prompting third-party safety audits and raising broader questions about agent visibility and data governance. Open article →