Claude Sets Automatic Mode as Default for Pro Max and Team Plans
Anthropic announced that starting next week, automatic mode will be enabled by default for Claude's web, desktop, and mobile sessions in the Pro, Max, and Team plans, eliminating the need for users to approve each step individually.
The system will now automatically check for high-risk behaviors, and users can switch back to manual mode at any time. This change applies to everyday chat and tool invocation scenarios, allowing sessions to run with fewer interruptions.
Previously, automatic mode became the default for the same plans in Claude Code in August 2026. Tests showed that the independent classifier intercepted 89% of dangerous commands, while manual approval only intercepted 14%. The classifier reviews potentially destructive actions, such as mass file deletions or sensitive data leaks, before each tool invocation.
Automatic mode aims to support longer autonomous tasks while reducing the risks associated with completely bypassing permissions. When the classifier intercepts, the model typically seeks a safer path or directly requests user confirmation. After a certain number of consecutive interceptions, it will automatically revert to manual mode.
This feature was previously optional for Enterprise and some cloud platforms, and is now being promoted as the default for a broader range of paid users. Users who have fixed other default permission settings can keep them unchanged or receive a one-time switch prompt.
In market dynamics, buyers are paid Claude users seeking to reduce approval friction to enhance productivity; sellers are Anthropic, increasing product stickiness and usage duration through default higher autonomy. This is a product feature-driven event, with funding and attention shifting from manual supervision to automated safety layers. Beneficiaries are heavy users and developers seeking efficiency, while those under pressure are users with strong demands for complete manual control or doubts about the accuracy of automatic interceptions.
Source: Public Information
ABAB AI Insight
Anthropic previously launched an automatic mode research preview in Claude Code in March 2026 and set it as the default for Pro, Max, and Team in August, simultaneously stopping additional token charges for the classifier. This same logic is now being extended to web, desktop, and mobile, continuing its path of replacing item-by-item manual confirmation with an independent safety classifier.
The capital path reflects a reduction in user interaction costs, promoting longer autonomous conversations and tool usage. The motivation is to enhance product stickiness and task completion rates under safe and controllable conditions, transitioning the model from an assistive tool to an independently advancing agent, while reducing user churn due to frequent confirmations.
This is similar to the early transition of autonomous driving from complete human intervention to graded automation, or operating systems moving from item-by-item permission pop-ups to application sandboxes and permission group management. It is currently in the expansion phase of AI agents moving from "strong human supervision" to "default autonomy + safety interception."
Essentially, this is a technological substitution. The mechanism lies in the independent classifier significantly outperforming manual item-by-item review in tests (89% vs. 14%), allowing the system to maintain risk interception capabilities while removing interaction bottlenecks; users defaulting to higher autonomy is due to the improved actual interception effectiveness and usage efficiency, thus shifting control from step-by-step confirmation to a results-oriented safety layer.
ABAB News · Cognitive Laws
Default autonomy is safer than item-by-item approval
Human supervision is often decorative; machine interception is substantive
The key to efficiency improvement is reducing confirmations rather than reducing risks