Microsoft will begin rolling out AI-powered phishing detection in Teams in early November 2026, part of a targeted release phase expected to conclude by mid-November. General availability worldwide is projected for early to mid-December.

The feature was first disclosed in a Microsoft 365 Roadmap entry and confirmed in a Microsoft 365 Message Center update. It uses conversation context to flag phishing, social engineering, impersonation attempts, and other deceptive intent in Teams messages, focusing on higher-risk scenarios such as external first-contact chats.

Unlike traditional phishing detection that relies on identifying malicious links or attachments, the new capability analyzes behavioral patterns in conversations to catch suspicious activity that lacks those classic indicators.

  • No admin action is required to enable the protection for most users
  • The system does not retain message content for AI model training or other secondary uses
  • Microsoft states the capability is used solely for security protection, not for productivity, summarization, advertising, or user monitoring

This addition builds on a string of Teams security improvements Microsoft has rolled out over the past year. In December, the company gave admins the ability to lock out external users via the Defender portal to curb cybercrime groups, including ransomware gangs, from abusing Teams for social engineering against employees. In August, a meeting protection policy began blocking automatically identified external bots from joining Teams meetings.

More recent changes include blurring QR codes sent by external senders to reduce fraud and phishing risk, letting users report suspicious guest invitations directly within Teams, and plans announced in early October to support third-party deepfake detection and impersonation protection tools in Teams meetings.

Together, these updates reflect Microsoft’s ongoing push to harden Teams against increasingly sophisticated social engineering campaigns that exploit chat-based trust rather than relying solely on malicious payloads.