Claude Code gets 'auto mode': AI now decides its own permissions

Anthropic on March 25, 2026 released a new feature in Claude Code called 'auto mode'. The change lets the AI tool decide on its own which permissions it needs to complete a task, without requiring the developer to approve each individual step.

Until now, Claude Code has asked users for approval at critical junctures. Auto mode reverses this: the model evaluates the context and decides for itself whether it needs file write access, network calls, or command execution. The goal is fewer interruptions and a more seamless workflow.

Anthropic emphasizes that users can still override the model and that the change does not remove safety guardrails. The company describes it as a balance between autonomy and control, with auto mode intended for experienced users who already trust the model's judgment.

On the same day, Anthropic reported a major service disruption that particularly affected Claude Opus 4.6. The company said it was working to resolve the issue without giving details about the root cause.

For CIOs and development teams already using Claude Code in production, auto mode is a concrete step toward more agentic AI. It reduces friction in long coding sessions, but also demands a higher standard of evaluation regarding what the model actually does between each prompt.