Anthropic introduces new setting in claude code 2.1.311 to allow AI agents to delete files without user approval

On October 28, 2026, The Register reported that Anthropic has introduced a new environment variable in the v2.1.311 version of Claude Code. This variable, named CLAUDE_CODE_EVEN_MORE_DANGEROUSLY_SKIP_PERMISSIONS, allows users to disable permission checks for AI agents when performing actions like deleting files. Daisy Hollman, an Anthropic technical staff member, explained that the previous setting did not approve all operations, as the system still blocked clearly dangerous commands, such as recursively deleting a user's home directory. The new variable enables users to further bypass these protections. Hollman confirmed the change is real, though it does not remove all safety measures, as administrators can still impose additional restrictions. The variable's name and absence from the update log led to some skepticism, but Hollman clarified it is not a joke.