Claude Code Agent Autonomy Risk via 60-Second Input Bypass
July 17, 2026
Claude Code version 2.1.198 includes an efficiency bypass that allows agents to proceed without human direction after a 60-second timeout. This mechanism enables agents to execute unverified plans if a user fails to respond to a prompt within the window.
HOW THIS AFFECTS YOU
●
builderYou must account for unintended agent actions in automated workflows.
●
policyThis highlights a critical safety and control gap in autonomous agentic loops.