Microsoft Teams Will Start Flagging Phishing Chats With AI in November
Microsoft says Teams will begin using AI to spot phishing, impersonation and deceptive intent in chats, focusing on first contact from outside the company, with a rollout starting in early November. Here is how it works and why it should be one layer of several.
Here Is Why a Smarter Chat Filter Still Needs Tighter External Access, Trained Staff and a Call-Back Rule Behind It
The message arrives in Teams from someone claiming to be with the help desk. There is no link and no attachment, just a friendly request to approve a sign-in prompt or install a support tool. Traditional filters look for bad links and files, so a message like this passes straight through. Microsoft is now trying to catch that kind of message by reading the conversation for intent.
This Pick is based on BleepingComputer’s report on Microsoft’s announcement.
What is changing
According to a Microsoft 365 Message Center update, Teams will begin using AI-powered, context-aware phishing detection in early November 2026. The targeted release phase is expected to end by mid-November, with general availability worldwide likely between early and mid-December. Microsoft says the feature looks for phishing, social engineering, impersonation and deceptive intent, including suspicious behavior that has no malicious link or file.
For most users it will focus on higher-risk situations, especially first-contact chats from people outside the organization. Admins will not need to turn it on. Microsoft says the detection system does not retain message content for model training or other secondary uses, and that it is used only for security protection, not for productivity features, advertising or user monitoring.
Part of a longer list
This is the latest in a series of Teams security changes. Microsoft has previously let admins block external users through the Defender portal, introduced a policy to automatically keep identified external bots out of meetings, announced blurring of QR codes from external senders and a way for users to report suspicious guest invitations, and said it will support third-party deepfake detection in meetings.
Why one filter is not enough
Detection based on intent is a real step forward, but no filter catches everything, and attackers adjust their wording quickly. The fake help desk chat works because it takes advantage of trust and urgency. A layered approach assumes some messages will get through and makes sure each one runs into another barrier:
- Narrow who can reach you. Review your external access settings so staff can only chat with outside organizations you work with, or require approval for new external contacts.
- Make outsiders obvious. Keep external and guest labels visible, and teach staff that real internal IT will not appear as an external contact.
- Use a call-back rule. Any request to approve a sign-in, share a code, install remote access software or change payment details gets verified through a number or channel you already know.
- Limit what one click can do. Block unapproved remote support tools from being installed, and use phishing-resistant multifactor sign-in for admins and finance staff.
- Make reporting easy. Show staff how to report a suspicious chat, and route those reports to someone who acts on them the same day.
What this means for your company
If you use Microsoft Teams, this feature will arrive without any work on your part, and it should catch some attempts that slip past today. Treat it as one layer. Use the next few weeks before the rollout to tighten external chat settings, remind staff how a real help desk contacts them, and confirm that your call-back rule covers chat as well as email and phone. Products mentioned here are examples. Whatever collaboration platform you use, the same layers apply.
Need help tightening external chat settings and building a layered defense against help desk impersonation in Teams? Reach out to Layer 7 Data Solutions to discuss how we’ve helped companies similar to yours.
About The Author
