Claude Code introduces 'auto mode' where an AI classifier (Claude Sonnet 4.6) reviews and approves actions before execution, blocking scope escalation, untrusted infrastructure targeting, or hostile content. The system includes default safety filters and customizable rules, with detailed JSON output showing allowed operations like local file actions within project scope and test artifacts. This represents a significant step toward safer autonomous AI coding assistance with built-in safeguards.
Background
Claude Code is an AI-powered coding assistant from Anthropic that helps developers with code generation and automation tasks. Previous versions required explicit permission flags like --dangerously-skip-permissions for autonomous operation.
- Source
- Simon Willison
- Published
- Mar 25, 2026 at 07:57 AM
- Score
- 6.0 / 10