OpenAI Agents Hit UN Site 16,000 Times: What Went Wrong

OpenAI agents scanned a UN data site more than 16,000 times and changed tactics when access failed, raising new questions about autonomous AI controls.