GitHub Launches Assisted Permissions to Automate Safe Copilot Actions

GitHub on October 26 rolled out "Assisted permissions," a security-focused authorization feature for its AI coding assistant, GitHub Copilot. The update utilizes large language models to assess the safety of operations requested by the tool, granting automated approval solely to actions that are clearly determined to be secure. The rollout targets developers seeking a balance between uninterrupted coding efficiency and safe machine-driven software management.

GitHub Copilot functions as an AI development utility capable of supporting programmers across various stages of software production. Its features encompass generating and completing code, answering programming questions, editing project files, and executing terminal commands. As software engineers increasingly rely on automated tools to accelerate complex engineering processes, governing the actions taken by such AI agents has become a focal point of development operations.

Prior to the introduction of Assisted permissions, developers were forced to rely on two extreme permission paradigms when managing Copilot's system interactions. Under the standard manual approval mechanism, every single non-permitted operation paused the development process to request direct confirmation from the user. While highly secure, this workflow repeatedly interrupted programmers. On the opposite end of the spectrum, the "Allow All" setting approved every command and file request indiscriminately, offering seamless execution at the cost of eliminating oversight entirely.

Assisted permissions is positioned squarely between these two extremes. When Copilot requests to execute commands, open or edit local files, or retrieve external web addresses via URLs, an integrated large language model inspects the nature of the request. The model assesses whether the proposed operation poses any operational risk. If the system verifies the action as explicitly safe, execution proceeds immediately without prompting the engineer.

Developers can access and configure the new permission tier across multiple programming environments supported by GitHub. Within the Copilot Command Line Interface (CLI), users can enable the feature by executing the command "/permissions" and choosing "assisted," or by typing "/permissions assisted" directly into the prompt. Developers working inside the GitHub Copilot application can configure the feature through the "Tool permissions" section on the mode selection screen, while programmers using Visual Studio Code can select it from the editor's permissions configuration panel.

Beyond interactive desktop use, the new feature has also been tailored for headless environments such as automated continuous integration and continuous deployment (CI/CD) pipelines and recurring background jobs. In these setups, human intervention is fundamentally unavailable to approve individual prompts. To support these scenarios, GitHub incorporated a dedicated non-interactive option in the Copilot CLI using the "--assisted-approval" command-line flag.

When running with the non-interactive option enabled, any Copilot request evaluated as safe is cleared automatically. If a requested operation fails the safety evaluation or requires explicit user verification, the request is immediately denied rather than left hanging. In addition to denying the action, the system passes an explanation of the rejection back to Copilot. Armed with this context, the AI assistant can analyze why the operation was blocked and independently attempt alternative, compliant strategies to accomplish the assigned task.