An auto mode will be enabled by default in Claude Code to improve the behavior of 'requiring human judgment for each command execution.'
Claude Code is an AI agent that performs coding and PC tasks on behalf of humans, but it requires human permission to execute commands, and its work will stop until permission is obtained. It has now been announced that an '
Auto mode is now the default in Claude Code for Pro, Max, and Team plans | Claude by Anthropic
https://claude.com/blog/auto-mode-default-in-claude-code
Running auto mode in production | Claude by Anthropic
https://claude.com/blog/auto-mode-in-production
Starting August 14, auto mode will be the default permission mode in Claude Code for Pro, Max, and Team users.
— ClaudeDevs (@ClaudeDevs) August 7, 2026
Auto mode reviews shell commands and actions with a separate classifier. In testing, it caught 89% of dangerous commands. Manual approval caught 14%.
According to Anthropic, users approve 97% of Claude Code's permission prompts. Anthropic states, 'Such a high approval rate suggests that many users are reflexively approving each command without reviewing it. Claude Code asks users to make dozens to hundreds of critical security decisions per day, and the burden of reviewing these decisions increases the likelihood that important commands will be overlooked,' highlighting the high risks of the traditional method of requiring users to manually review each command.
Anthropic conducted a test with 1053 paid testers, in which they performed a research task using Claude Code. During the test, one of the 'user approval pending' prompts was replaced with one containing a dangerous command, but testers were only able to detect the dangerous command in 13.6% of cases. In contrast, the auto mode was able to detect and block 89% of the dangerous commands.
The reliability of the auto mode has also been confirmed in tests conducted by the third-party organization Trajectory Labs. The following figure shows the success rate of 72 types of prompt injection attacks attempted on 'Claude Sonnet 5 (Claude Code)', 'Claude Fable 5 (Claude Code)', 'Claude Opus 5 (Claude Code)', and 'GPT-5.6 Sol (Codex)'. When GPT-5.6 Sol was running on Codex with automatic approval mode enabled, 5.83% of the attacks succeeded, but Claude Code's auto mode prevented all attacks.
Claude Code's auto mode will be enabled by default for Pro, Max, and Team plan users starting August 14, 2026. For Enterprise plan users and those using the API, it is scheduled to be enabled from September 2026 onwards.
Related Posts:
in AI, Posted by log1o_hf