Claude Code Makes Auto Mode Default: Does It Solve Coding Agent Security?

Auto mode is a setting in Claude Code that allows the agent to execute actions automatically without human approval for each step, relying on built-in risk assessments.

According to Anthropic's tests, humans approved 86.4% of harmful actions in their trials, while auto mode blocked 89% of them, suggesting it is superior in preventing errors.

Anthropic claims third-party tests showed zero successful attacks out of 720 indirect prompt injection scenarios, but experts demand independent confirmation and point to potential vulnerabilities.

It is advisable to adopt it cautiously, reviewing security policies and testing in non-critical environments first, given the lack of sufficient independent confirmation.