TechCrunch: Anthropic gives Claude Code an auto mode with a built-in safety layer

Anthropic has released auto mode for Claude Code in research preview, giving the AI the ability to decide independently which actions are safe to take without waiting for user approval at each step. The feature builds on Claude Code’s existing “dangerously-skip-permissions” command — which previously handed all decision-making to the AI with no guardrails — by adding an AI safety layer that reviews each proposed action before it runs.

The mode currently works with Claude Sonnet 4.6 and Claude Opus 4.6. Anthropic recommends using it in sandboxed, isolated environments to limit the potential impact if something goes wrong. The company has not published the specific criteria its safety layer uses to distinguish safe actions from risky ones. Auto mode is rolling out to Enterprise and API users.

The release comes alongside two other Claude Code updates: Claude Code Review, an automatic code reviewer designed to catch bugs before they reach the codebase, and Dispatch for Cowork, which routes tasks to AI agents that handle them asynchronously.

For product managers involved in AI tooling decisions or working with teams that use Claude Code, the practical question auto mode raises is about trust calibration: how much autonomous action is acceptable in a given context, and what the right boundaries are. The research preview framing signals that Anthropic is still gathering evidence on where those lines should sit.

The direction across the industry is consistent. GitHub and OpenAI have both shipped autonomous coding tools that act without waiting for human approval at every step. Auto mode fits that pattern while adding an explicit safety review layer — a middle position between full human oversight and unchecked autonomy.