Anthropic will switch Claude Code auto mode to the default setting for Pro, Max and Team accounts from Aug. 14. TechCrunch reported on Saturday that Claude Code will no longer ask for human approval at every step. It will request approval only when it judges a task to be irreversible, destructive or aimed outside the user environment.

Anthropic first unveiled auto mode as a beta in March. At the time, the feature was presented as a way to balance speed and control.

Anthropic said tests showed auto mode delivered safer results than manual review. In a study of 1,053 paid testers, auto mode caught 89 percent of harmful behavior, while human manual review filtered out only 13.6 percent. Users approved 97 percent of Claude Code permission request prompts, it said.

Boris Cherny (보리스 체르니), head of Claude Code, said in a post on social media X (Twitter) that his team and he have been using only auto mode for months. He also said it is hard to go back to the permission request prompts.

Anthropic also said it is adding safety features such as prompt injection checks and custom enforced refusal rules to prevent actions such as data exfiltration.