Researchers Uncover New Phishing Risk Hidden Inside Microsoft Copilot

### Executive summary Research by Permiso demonstrates a realistic proof-of-concept where attacker-controlled text in emails can manipulate Microsoft Copilot summaries via cross-prompt injection, causing the assistant to present phishing-style alerts or malicious prompts inside the trusted AI interface; tests showed varying behavior across Outlook and Teams and the report recommends patches, access restrictions, email filtering, monitoring, and user training to mitigate this AI-assisted phishing vector.