Execution safety for AI agents that take real actions
Execution safety for AI agents that take real actions.