Anthropic now lets Claude users jump out of plane without parachute
Anthropic has added an option in Claude Code v2.1.311 that lets developers bypass approval prompts for commands such as `rm`, which can delete files. The change removes a safeguard that remained even when users enabled the tool’s existing permissions bypass, increasing the risk of accidental data loss.
The option is activated by setting the environment variable `CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS=1`. Anthropic staff member Daisy Hollman confirmed the feature and said it disables the block on commands such as `rm -rf ~`; managed settings may still restrict it. Claude Code has several permission modes, and its “auto” mode became the default in August, while bypass mode runs without background safety checks.
In short
- Claude Code can now run file deletion commands without approval.
- The new setting is an environment variable in version 2.1.311.
- Anthropic warns through the setting’s name that it is risky.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Developers who knowingly enable this option may value speed and control, especially in trusted, isolated environments or automated workflows where repeated approval prompts interrupt work. Making the setting explicit and unusually conspicuous can be seen as giving experienced users a choice, while managed settings let organisations retain restrictions.
The case against
Removing the remaining prompt for destructive commands makes a serious mistake easier, including accidental deletion of important files, and users may not fully anticipate how an AI-generated command will behave. Critics can argue that convenience does not justify weakening a safeguard that helps catch errors, particularly when less risky ways to reduce interruptions may be available.