Vulnerability / research

Microsoft demonstrates how a phished authorization can expose enabled coding-agent features

Reported

Published here

What happened

In a controlled simulation, a test user approved an attacker-initiated GitHub device code. Researchers then explored access to enabled Copilot CLI features and resources available to that account.

Parties and reporting

Where this happened

Victim location not disclosed. Real organizations or systems were affected or targeted, but the sources do not say where. Explore the world map.

AI tools and systems: GitHub Copilot CLI, OAuth device authorization, Copilot Memory

Outcome

The lab demonstrated remote-session steering, misleading memory entries, synchronized-session reading and private-repository access. Each depended on permissions, requested scopes and enabled features; local command approval still applied.

Operating lesson

Approve only sign-ins you initiate, restrict remote-session and memory features, and revoke unexpected OAuth authorizations.

Evidence and limits

Microsoft explicitly describes existing capabilities working as designed, not observed attacks or unauthenticated code execution. This research entry does not assert a separate software defect or affected customer.

Sources