Critical CoPhish Exploit Uses Copilot Studio to Hijack OAuth Tokens

Security researchers at Datadog have uncovered a sophisticated phishing technique dubbed “CoPhish” that weaponises Microsoft Copilot Studio to exfiltrate OAuth tokens. The attack exploits the ability of Copilot Studio agents to create configurable “topics” and use a built‑in Login button that can redirect users to OAuth phishing pages, with tokens later sent to attacker‑controlled servers.

It relies on the trusted appearance of official Microsoft domains, including copilotstudio.microsoft[.]com, to persuade users to consent to malicious applications. In July 2025, Microsoft introduced a default application consent policy that blocks high‑risk permissions, though some permissions such as email and calendars remain consentable, and later in October 2025 Microsoft planned to update the default policy further.

The technique can allow attackers to backdoor a sign‑in topic and forward tokens server‑side via an HTTP request action, enabling access to a user’s email and other data depending on the permissions granted. The article recommends stronger application consent policies and monitoring of Entra ID audit logs to detect suspicious consent activities.