Claude Code Defaulting to Auto Mode, AI Takes Over Security Approvals

Anthropic has announced that Claude Code will default to Auto Mode starting August 14, with new sessions for Pro, Max, and Team users automatically using this mode. This mode uses a safety classifier to automatically determine whether each action is executable, reducing the frequency of manual confirmations by users. Official tests show that human approval is not very effective; among 1,053 paid professional testers, only 13.6% of clearly dangerous commands were intercepted, while Auto Mode achieved an interception rate of 89%. After processing 50 permission pop-ups in succession, the human interception rate dropped to about 5%. In actual use, Claude Code users approve 97% of permission requests, and a quarter of interaction sessions directly close permission checks. Anthropic points out that frequent pop-ups have led users to develop a habit of "mindlessly clicking agree." Auto Mode will automatically allow safe operations while intercepting high-risk actions such as deleting data and sending information externally. Although each judgment in Auto Mode consumes a small amount of tokens, Pro, Max, and Team users do not have to pay for it. Enterprise, API, and major cloud platform users still need to manually enable this mode for now, but there are plans to change it to a default setting in the future.