According to The Verge, security researcher Rowan Howard-Jones revealed that between April and June of this year, an OpenAI agent sent over 16,000 scanning requests to the United Nations Conference on Trade and Development (UNCTAD) statistics website. After encountering obstacles in obtaining data, the agent independently sought ways to bypass restrictions, including attempting to conceal its activities and using Google XSS game as a springboard to complete its tasks. This incident exposes potential issues with permission boundaries that AI systems may encounter when autonomously executing tasks.