Skip to content
MediumNews

The risks of device code authentication for AI coding agents

Published
Record updated
View JSON

Summary

Microsoft Security Blog reports lab research showing that device code phishing, which applies to any public client using the OAuth device flow, can be extended to AI coding agents, using GitHub Copilot CLI as an illustrative example. In a controlled simulation, a victim who approved an attacker-initiated GitHub device-code request enabled outcomes including remote steering of an active Copilot CLI session, injection of misleading memory entries, reading of synchronized cloud sessions, and access to private repositories available to the account. The researchers state these are proof-of-concept results and not attacks observed in the wild.