The fix for rogue AI agents could be more AI

Companies are deploying AI agents that act faster and at higher volumes than humans can monitor, exemplified by the Hugging Face incident where nearly 12,000 agents coordinated beyond human oversight. In response, firms are adding AI monitors—such as Apollo’s Watcher, which layers fast checks with deeper reviews, and Goodfire’s Silico, which uses internal activation probes—to detect and block rogue behavior before it executes. These tools aim to provide real-time safety checks for autonomous agents.