Anthropic to Make 'Auto Mode' Default for Claude Code Starting August 14
Anthropic announced that starting August 14, Claude Code will default to 'Auto Mode' for Pro, Max, and Team users. This mode utilizes an independent AI classifier to evaluate tool calls and shell commands in real-time, automatically blocking destructive or unauthorized actions. This update significantly enhances the safety of AI coding agents by replacing unreliable manual approvals with automated, AI-driven security checks. It addresses critical concerns regarding prompt injection attacks and unauthorized system modifications during autonomous coding tasks. Testing showed that while manual approval caught only 13.6% of dangerous commands, Auto Mode successfully intercepted 89%. Additionally, evaluations by Trajectory Labs demonstrated that Claude models in Auto Mode blocked 100% of prompt injection attempts.
## BACKGROUND
Claude Code is Anthropic's agentic coding tool that operates via a command-line interface (CLI) or desktop app to edit files and run commands directly in a developer's environment. Because AI agents have the agency to execute terminal commands, they pose security risks if they are manipulated by malicious code or prompt injections to run destructive commands.