Microsoft adds context-based phishing screening to Teams chats
Microsoft will start testing an AI feature in Teams in early November 2026 that examines conversation context for phishing, impersonation, and social-engineering signals, including messages from new external contacts. The company says the system will not keep message content or use it for model training, advertising, or monitoring. The capability is expected to reach general availability around December, and administrators will not need to enable it.
Microsoft's rollout will first go to a targeted release group in early November 2026, finishing around mid-month, before a broader launch expected between early and mid-December. The company outlined the change through its Microsoft 365 Roadmap and a Message Center notice, saying administrators will not have to switch it on.
The screening is aimed mainly at first conversations with outside contacts, where harmful intent may lack obvious links or attachments. It joins other Teams safeguards, including controls for external users, automatic blocking of identified outside bots in meetings, QR-code blurring, guest-invitation reporting, and planned deepfake and impersonation protections.
Teams users, especially employees who receive messages from unfamiliar outside contacts, could benefit from earlier warnings about impersonation and social-engineering attempts. Administrators may see fewer successful phishing incidents without adding configuration work, while privacy-sensitive organizations may welcome the stated limits on content retention. Because the screening focuses on context rather than only links or files, it could catch some deceptive messages that conventional filters miss, though its real-world effect will depend on accuracy and user trust.