Skip to content
Claude News

Appearance

Appearance

Claude Code

RuntimeWire: Claude Code puts staff bypass in Accept edits

Written by an AI model from the linked sources and checked against them by a second model. How we work

When an Anthropic employee’s session asks Claude Code for bypass permissions mode, the client starts it in Accept edits instead. That is what RuntimeWire found when it reverse engineered the tool’s publicly distributed JavaScript on October 18. The behaviour sits in a separate session-start branch for employee accounts.

Anthropic’s permission-mode documentation says bypassPermissions lets Claude take every action without asking first, and describes it as suited to “Isolated containers and VMs only”. acceptEdits covers reads, file edits and common filesystem commands such as mkdir, touch, mv and cp.

RuntimeWire also found a second path in the same client. The API-error parser files refusals of staff-linked sessions that run unapproved security controls under their own category, separate from errors that flag an outdated client, and each category maps to its own staff-facing message. In the outlet’s words, the code shows how the client “classifies and explains these conditions”.

The work was done on the client alone. RuntimeWire traced the parser’s error categories to their messages, examined the employee session-start branch, and kept the retrieved files’ bytes along with their SHA-256 hashes. The server rules that decide when a staff session gets refused are still unknown, and so is whether customers have access to equivalent controls.

RuntimeWire observed no live API response and was not able to independently reproduce the finding.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.