Test your understanding of security controls, sandboxing, policy evaluation, and privilege boundaries in production autonomous AI agent deployments.
In a classical Reasoning and Acting (ReAct) loop, how does state corruption occur when an agent parses an unexpected tool execution output?