Claude Code: Anatomy of a Misfeature (www.olafalders.com)

On July 1, 2026, Anthropic released an unexpected feature in Claude Code version 2.1.198 that enabled an automatic decision-making process after a 60-second inactivity period. This "auto-continuation" allowed the AI to proceed without user input, which raised several concerns, particularly around trust and control. As users continued to run multiple instances of the AI, the feature could lead to unintended actions being taken without explicit user directives, posing risks within automated deployments. The significance of this event lies in its implications for AI governance and user trust. Many users expressed frustration that such a critical change was not documented in the changelog, highlighting potential gaps in software transparency. Following widespread complaints, Anthropic addressed the issue by reversing the default behavior in version 2.1.200 to prevent auto-continuation unless explicitly enabled. While this fix resolved immediate user concerns, it left open questions regarding the oversight processes in place at Anthropic, particularly about how such a feature was developed and deployed without proper checks or documentation. This incident underscores the need for better communication and oversight in AI feature releases to foster user confidence and accountability.