Claude Code 2.1.198: The 60-Second Auto-Proceed Misfeature
Original: Claude Code: Anatomy of a Misfeature
Why This Matters
Undocumented autonomous behavior in widely-used AI coding agents raises serious questions about release governance and human oversight.
On July 1, 2026, Anthropic shipped Claude Code v2.1.198, which introduced an undocumented 60-second auto-proceed timer. If a user fails to respond to an agent prompt within that window, Claude Code continues autonomously — a behavior never mentioned in the changelog.
Claude Code version 2.1.198, released on Canada Day (July 1, 2026), included an undocumented 'efficiency bypass': if a user does not respond to an agent prompt within 60 seconds, the agent proceeds on its own judgment and logs 'No response after 60s — continued without an answer.' Blogger Olaf Alders documented the behavior firsthand, noting that the session literally printed 'The user stepped away. I'll proceed with best judgment.' The change was not mentioned in any changelog or release notes; documentation was only updated after the fact. A fix was shipped within a couple of days, but Alders raises unresolved questions: whether the feature was designed, coded, reviewed, and released without meaningful human oversight, and whether Claude Code's auto-update mechanism itself constitutes a security and trust risk. He also notes that disabling auto-update as a workaround also prevents plugin updates. The incident highlights risks around agentic AI tools operating in deployment or production contexts, where an autonomous wrong decision could have significant consequences.