Anthropic sets Claude Code auto mode as the default for paid accounts

Anthropic will make auto mode the default setting in Claude Code for Pro, Max and Team accounts beginning August 14. The change reduces the number of approval prompts users see while working with the coding tool, allowing Claude Code to continue on its own in many situations.

Under auto mode, Claude Code will proceed unless it judges an action to be irreversible, destructive or directed outside the user’s environment. Anthropic first introduced a test version of the mode in March, presenting it as a way to balance speed with user control.

From approval prompts to risk-based intervention

The default shift reflects Anthropic’s view that repeated manual permissions do not necessarily produce better oversight. In a study involving 1,053 paid testers, the company said auto mode identified 89% of harmful actions. Human review identified 13.6%.

Anthropic said manual review can become habitual, noting that Claude Code users approve 97% of permission prompts. Rather than ask for confirmation at every step, auto mode is designed to reserve intervention for actions that cross its stated risk thresholds.

Claude Code Head Boris Cherny said that he and the team have used auto mode exclusively for many months. The wider change means paid-account customers will encounter this operating model by default rather than as an optional test feature.

Safety controls remain central to deployment

Anthropic said it has also introduced safety capabilities including prompt-injection screening and customizable hard-deny rules. The hard-deny rules are intended to prevent specific categories of activity, including data exfiltration, while prompt-injection screening addresses instructions that may attempt to manipulate an AI system’s behaviour.

The default automation approach follows Anthropic’s broader effort to put its products into persistent workplace workflows, including Claude Tag’s always-on Slack AI colleague, which brings an always-on AI colleague with memory into Slack. Both product directions place more emphasis on defining operational boundaries than on seeking confirmation for every routine action.

For businesses using Claude Code, the practical implication is to establish and test hard-deny rules, clarify which actions can leave the working environment, and evaluate how auto mode handles their own development workflows before relying on fewer permission prompts.