Anthropic previews auto mode for Claude Code to approve low-risk actions

In short: Anthropic is testing an “auto mode” in Claude Code that lets the AI approve some low-risk coding actions without asking every time.

What happened

Anthropic launched a research preview of “auto mode” for Claude Code, a tool that helps developers work with code. The new mode lets Claude decide when it is safe to proceed, so it can handle routine permissions during a coding session.

In practice, this means Claude can approve lower-risk actions, like editing certain files or running some commands, without constantly interrupting the person using it. For higher-risk actions, it still stops and asks, similar to a car that can handle easy stretches of highway but alerts you before a tricky merge (it can help, but you are still responsible).

Anthropic says this is meant to be safer than what some developers do today, which is turning off permission checks entirely to avoid repeated prompts. Auto mode can be enabled from the command line using claude --enable-auto-mode.

The preview also includes “prompt injection safeguards,” which are protections against hidden instructions that could be buried in files or tool output (like a note that tries to trick the AI into doing something harmful). Organizations can disable auto mode across managed devices using admin controls, such as MDM or operating system policies.

Anthropic notes tradeoffs. Auto mode can use more tokens (the chunks of text an AI reads and writes), which can increase cost and slow responses. The company recommends using isolated environments when possible.

Why it matters

AI coding tools are getting more independent, and that can save time. But it also raises safety and cost questions, especially for teams working on important systems. Auto mode shows how vendors are trying to balance speed with guardrails, rather than asking users to choose between convenience and control.

Source: TechCrunch AI