AI Agents Must Be Monitored in Real Time, Warn Partnership on AI Experts

Madhulika Srikumar and Claire Leibowicz of the Partnership on AI, a nonprofit focused on AI governance, are calling for real-time monitoring of AI agents. They argue that as autonomous systems become more capable, tracking their actions is essential to prevent harmful outcomes and ensure accountability. Without such oversight, AI agents could make decisions that are difficult to trace or correct, posing risks to individuals and society. The authors emphasize that current monitoring methods are insufficient for the speed and complexity of modern AI. They propose a framework for continuous observation and evaluation, enabling rapid intervention when needed. This approach would help build trust in AI systems and align their behavior with human values. The need for real-time monitoring is urgent as AI agents are increasingly deployed in high-stakes environments. The authors conclude that proactive oversight is not just a technical challenge but a societal imperative.