The vulnerability in livebook-dev livebook involves improper origin validation in the handling of keyboard events. Livebook renders untrusted notebook JavaScript inside a sandboxed cross-origin iframe to isolate it. However, the trusted iframe shell forwards all keydown events to the parent page without verifying if the event is genuinely user-generated (Event.isTrusted). The parent page then re-dispatches these events without validation, allowing untrusted scripts to simulate keyboard shortcuts. Some shortcuts trigger immediate actions without confirmation, such as full evaluation of all notebook cells and runtime restart, which can lead to arbitrary code execution. This affects livebook versions from 0.5.0 up to but not including 0.18.7, and from 0.19.0 up to but not including 0.19.9.

An attacker can cause a user's livebook session to execute all notebook cells automatically and restart the runtime without user consent by injecting malicious JavaScript in notebook outputs. This can lead to execution of arbitrary Elixir code on the user's runtime environment. Additionally, because outputs are mirrored to all connected clients, collaborators can be affected simultaneously. The forced full evaluation and runtime restart can disrupt workflows and potentially execute harmful code.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, users should avoid opening notebooks from untrusted sources or containing untrusted JavaScript outputs. Monitoring vendor channels for updates and applying patches promptly once released is recommended.