Microsoft Copilot Studio agents exploited in new phishing campaign

Microsoft has confirmed plans to address a newly discovered phishing method known as CoPhish, which abuses Copilot Studio agents to send fraudulent OAuth consent requests through trusted Microsoft domains, according to BleepingComputer.

Researchers at Datadog Security Labs discovered the CoPhish campaign, which exploits Microsoft Copilot Studios chatbot feature to create phishing schemes that appear to come from trusted Microsoft domains. Attackers use the platforms demo website option to host fake login pages that send fraudulent OAuth consent requests, tricking users into authorizing malicious applications.

Datadog researcher Katie Knowles explained that attackers can design login buttons connected to rogue apps, enabling them to steal session tokens once a user signs in. Microsoft said upcoming policy changes will reduce risks for unprivileged users but warned that administrators with elevated permissions could still be targeted.

To mitigate attacks, both Microsoft and Datadog advised limiting admin privileges, tightening app permissions, and monitoring Copilot Studio and Entra ID activities for suspicious consent behavior.