2026-11-17 · Microsoft Security · MEDIUM

Device Code Phishing Threatens AI Coding Agents

60-WORD SECURITY BRIEF

Lab research shows device code phishing can target AI coding agents, using GitHub Copilot CLI as a proof-of-concept. The technique exploits weak consent verification during device authorization, allowing attackers to capture user credentials and potentially run arbitrary code. Teams should enforce multi-factor authentication for device flows, audit logs, and restrict AI agent permissions to mitigate credential theft.