Anthropic has made auto mode the default permission setting in Claude Code for Pro, Max, and Team plans, with the change taking effect for new sessions starting August 14, 2026.
According to the official Anthropic blog post published August 7, users on those plans who have not previously set a different default will begin new sessions in auto mode. Those who already configured a different default may receive a one-time in-product prompt asking whether they want to switch. Any pinned defaults remain unchanged.
How Auto Mode Works
In auto mode, each tool call is routed through a safety classifier before execution. The classifier is designed to block actions that are irreversible, destructive, or directed outside the user’s environment. If a call is blocked, Claude attempts a safer alternative or requests explicit approval. After three consecutive blocks or 20 blocks within a single session, the system falls back to traditional manual approvals.
Permission rules continue to apply before the classifier, with the exception of broad allow rules (such as unrestricted Python execution), which are set aside while auto mode is active. User settings files are left unmodified.
Safety and Performance Data
Anthropic released several quantitative results supporting the change:
- In a controlled study of 1,053 paid testers, human reviewers caught only 13.6% of deliberately inserted dangerous commands (143 of 1,053). Auto mode blocked 89% (937 of 1,053). Auto mode blocked 800 commands that humans approved; humans blocked only 6 that auto mode allowed. Human detection rates also declined over longer sessions (from ~17% early to ~5% after 50+ prompts), while auto mode performance remained consistent.
- In real-world production sessions (May–June 2026 data from opted-in accounts), manual approval sessions showed serious unintended harm roughly twice as often as auto mode sessions. At higher severity levels (7+ on a 0–10 scale), the rates were 6.3% for manual versus 2.4% for auto mode.
- Independent prompt-injection testing by Trajectory Labs (720 attempts across 72 scenarios) recorded a 0% success rate against Claude Fable 5, Opus 5, and Sonnet 5 running in auto mode.
Anthropic also noted that auto mode users on Team and Enterprise accounts have shipped approximately 25% more pull requests in production settings.
Billing and Rollout Scope
Effective immediately with the announcement, Anthropic is no longer charging Pro, Max, and Team users for the additional tokens consumed by the auto mode classifier. The company plans to extend the no-charge policy and the default setting to Claude Enterprise, the Claude API, Claude Platform on AWS, Google Cloud’s Agent Platform, and Microsoft Foundry within the coming month. Enterprise administrators can already set auto mode as the organization-wide default through managed settings.
Anthropic stated that auto mode reduces risk relative to habitual manual approvals but does not eliminate it, and recommended continued human review for high-stakes production changes.
Users can toggle modes with Shift+Tab in the CLI or via the permission dropdown in the desktop application.