Claude Code Auto Mode: Smarter Permission Handling Without Skipping Safety
Claude CodeAnthropic introduces Auto Mode for Claude Code, a new permission setting that uses a classifier to automatically approve safe tool calls while blocking destructive actions — positioning itself between the default ask-every-time mode and the risky --dangerously-skip-permissions flag. Available as a research preview on Team plans.
Sources & Mentions
5 external resources covering this update
Auto mode for Claude Code (Official Blog)
ClaudeLog
Auto mode for Claude Code — Simon Willison's Blog
Simon Willison's Blog
Claude Code gives developers 'auto mode,' a safer alternative to skipping permissions — 9to5Mac
Link
Claude Code AI tool getting auto mode — InfoWorld
Link
Claude Code gets auto mode to reduce interruptions — Techzine
Techzine
What Changed
Claude Code now ships Auto Mode, a third permission tier that lets the agent make permission decisions on your behalf. Before each tool call executes, a safety classifier reviews it for potentially destructive patterns — mass file deletions, sensitive data exfiltration, malicious code execution — and either lets it through or blocks it and redirects Claude toward a safer alternative.
Why It Matters
Until now, developers faced a binary trade-off: either approve every file write and bash command manually (safe but slow), or pass --dangerously-skip-permissions to remove all guardrails (fast but risky). Auto Mode fills the gap — fewer interruptions than default, less risk than skipping permissions entirely.
How It Works
- A classifier evaluates each tool call before execution
- Safe actions proceed automatically
- Risky actions are blocked; Claude is redirected to an alternative approach
- If Claude repeatedly attempts blocked actions, a user permission prompt is triggered as a fallback
- Compatible with Claude Sonnet 4.6 and Opus 4.6
Availability
- Research preview on the Team plan as of March 24, 2026
- Enterprise plan and API users: rolling out in the coming days
Caveats
- Auto Mode may still allow some risky actions when user intent is ambiguous or environmental context is insufficient
- May occasionally block benign actions
- Minor increases in token consumption, cost, and latency are expected
- Anthropic still recommends using Claude Code in isolated environments for maximum safety