The latest in managing and auditing GitHub Copilot agents

Quick Overview

GitHub Copilot Enterprise now offers expanded management and auditing capabilities, allowing administrators to create granular custom enterprise roles specifically for AI controls, define agent configuration rules using code rulesets that apply enterprise-wide, and track all AI agent activities via an audit log filtered by actor:Copilot.

Key Points: GitHub introduced new tools for controlling Copilot data and access, including expanding agent configuration and access control capabilities for enterprises. Administrators can create custom enterprise roles, such as 'AI Admins' (00:54), and assign specific permissions like 'Manage enterprise AI controls' (01:01). Custom AI agents, defined via Markdown files in a repository like 'github-private/agents' (01:39), can be made available across the entire enterprise by selecting the source organization (01:56). Enterprise-level push rulesets can be created under Policies (02:05) to restrict changes to agent configuration files, such as limiting pushes that modify files in the 'agents//.md' path (02:19). The Audit log automatically filters events where the actor is 'Copilot' (02:36), showing detailed JSON event data for actions like pull request creation (02:50). The Model Context Protocol (MCP) settings allow administrators to enable MCP servers everywhere or restrict access to only those listed in the enterprise's registry (03:36, 03:47).

Context: This video details recent enhancements to GitHub Copilot's enterprise management features, focusing on granular control over AI agents, Copilot usage, and security auditing. The demonstration moves through the Copilot Insights dashboard, enterprise role management, AI Controls for agents, Policy settings for rulesets, and finally, the Audit Log, illustrating how administrators gain control over Copilot's integration and operation within a large organization.

Detailed Analysis

The presentation showcases significant updates to GitHub Copilot Enterprise management and control features. It begins by reviewing the Copilot Usage Insights dashboard (00:00), which displays metrics like 508 active users and 82% agent adoption. The focus quickly shifts to governance, demonstrating how administrators can create custom enterprise roles, such as 'AI Admins' (00:49), and assign them specific permissions, like managing enterprise AI controls (00:59). Next, the video covers the creation of custom AI agents using configuration files stored in a repository (01:36), which can then be enabled enterprise-wide via the AI Controls Agent settings (01:56). A crucial governance feature shown is the ability to create custom Code Rulesets under Policies (02:05); specifically, a push ruleset is configured to restrict file modifications within the agent configuration directory, such as blocking changes to 'agents//.md' files (02:19). Finally, the demonstration moves to the Audit Log (02:32), where filtering by 'actor:Copilot' reveals detailed JSON logs of every AI action taken, including pull request creations (02:50), confirming who performed what action and where. The video concludes by reviewing the MCP settings, allowing admins to control whether developers can use external MCP servers or restrict them to only those listed in the enterprise's registry (03:36).