Human-in-the-Loop Failure

Claude Code's new default is a confession: the approval prompt was never a control

Anthropic's own study found users approved 97% of prompts and caught 13.6% of harmful actions. From August 14 the click stops being the safeguard, and deny rules become the job.