Anthropic now lets Claude users jump out of plane without parachute

Anthropic’s Claude Code v2.1.311 adds an environment variable that lets developers disable the remaining permission prompts for potentially destructive actions. The variable, CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1, goes beyond the existing --dangerously-skip-permissions option: that option bypasses most approvals but still blocks certain commands, including rm and rmdir, from running without confirmation. Anthropic staff member Daisy Hollman announced the change and confirmed it was real after users questioned it, in part because the variable’s name sounds deliberately alarming. The update was not listed in the release notes.

Claude Code offers several permission modes. Its auto mode allows actions while retaining background safety checks, whereas bypassPermissions removes those checks; until now, even bypass mode retained a prompt for some file-deletion commands. The new setting removes that final safeguard, though organization-managed settings may still impose restrictions. The change gives developers more control and fewer interruptions, but raises the risk of an agent carrying out destructive commands without a chance for human review. AI agents have previously deleted files or caused data loss through commands such as rm -rf ~, making the new opt-out especially consequential.