Claude Code Replaces Manual Permission Approvals With Intelligent Classifier

Anthropic's new classifier model in Claude Code automatically approves safe operations while blocking risky ones based on context, eliminating the tedious manual approval loop that prompted many developers to bypass security safeguards. The classifier evaluates scope, infrastructure trust, and prompt injection risks in real time, running on Sonnet 4.6 for speed and cost efficiency. This shift addresses a critical friction point that harmed both developer experience and security adoption.