Anthropic has introduced a new environment variable in version 2.1.311 of its Claude Code tool. This update allows AI agents to perform file deletions without requesting user approval for each action.
The new variable is named CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS. When activated, it removes the remaining safety filters that previously blocked high-risk commands, such as recursive deletion of home directory contents.
Daisy Hollman, a technical staff member at Anthropic, confirmed the feature is not a joke. The previous setting, dangerously-skip-permissions, still restricted certain critical operations, but the new variable disables these final safeguards.
The update does not remove security measures by default, as users must manually configure the environment variable. Additionally, any restrictions imposed by organization administrators remain in effect.
