Microsoft Teams will start using AI to flag phishing in chats from outsiders in November

Microsoft says Teams will begin using AI to spot phishing, impersonation and other social engineering in chats, starting with first-contact conversations from people outside your organisation. Targeted release begins in early November 2026 and worldwide availability is expected in early to mid-December 2026. No admin action is needed to turn it on, but the weeks before the worldwide rollout are a good moment to tighten Teams external access and refresh staff guidance on "IT support" messages.

What happened, in plain English

Microsoft Teams lets people chat not only with colleagues but, by default, with users in other organisations. That is handy for working with suppliers and customers. It also means a stranger with a Teams account somewhere else can send your staff a message.

Criminals have noticed. A common trick is to pose as the company's IT helpdesk ("we're updating your spam filter", "please verify your account") and talk the person into sharing their screen or starting a remote support session. In a report published on September 2, 2026, Microsoft Threat Intelligence described exactly this kind of campaign: attackers contacted users on Teams from a separate tenant (another organisation's Microsoft 365 environment), persuaded them to open Quick Assist, Windows' built-in remote help tool, and then installed malware.

The hard part for defenders is that these messages often contain no malicious link or file. Traditional filters look for bad links and attachments, so a polite, well-written message asking for help simply passes.

On October 21, 2026, Microsoft announced in its Microsoft 365 Message Center (notice MC1519866) that Teams will add "AI-powered, context-aware" detection. Instead of only scanning for bad links, it will look at what the conversation is trying to do: is this a stranger pretending to be someone they are not, or pushing for urgent action? If so, users may see a warning.

Think of it as a receptionist who does not just check visitors' bags but also notices when someone claiming to be "from IT" doesn't know anyone's name.

Are you affected? What to do now

Who it affects: organisations using Microsoft Teams in the worldwide multi-tenant cloud, on desktop, web, Mac, iOS and Android, according to Microsoft's roadmap entry (ID 575288). For most users, detection focuses on chats from external contacts in a first-contact conversation. Microsoft says Teams mobile users in China may also get extra protection for conversations inside their own organisation.

Timeline (from MC1519866):

• Targeted release: starts early November 2026, expected to complete by mid-November 2026.
• General availability worldwide: starts early December 2026, expected to complete by mid-December 2026.
• During an initial quality-validation period, detection may run in the background without showing warnings.

What admins should do:

• Nothing to switch on. Microsoft says no admin action or new configuration is required.
• Tell your privacy or compliance team. Microsoft states the change alters how customer data is processed, because Teams will analyse conversation context with AI/ML. It also states the detection is used only for security, not for productivity, summarisation, advertising or user monitoring, and that message content is not retained for model training or other secondary uses. Some organisations will want to record this in their data-processing documentation.
• Review Teams external access (Teams admin center → Users → External access). By default, users can chat with any external domain. Microsoft's own September 2 report recommends restricting external access to trusted domains. Options include an allow list of partner domains, blocking specific domains, and controlling whether unmanaged (personal) Teams accounts can start conversations with your staff.
• Check that users can report. In the Teams admin center, the messaging policy setting "Report a security concern" is on by default; confirm nobody has switched it off. Reports appear in the Microsoft Defender portal under Submissions → User reported (this requires Defender for Office 365 Plan 2 or Defender XDR). Microsoft asks admins to remind users to report suspicious activity and incorrect detections.
• Control remote support tools. Decide which remote assistance tools are allowed, and require identity or device checks before any remote session, as Microsoft's threat report recommends.
• Note the companion QR-code check. In a separate notice published on October 26, 2026 (MC1526651), Microsoft says Teams will also detect malicious QR codes in chats and channels, for internal and external messages, with worldwide availability from early to mid-November 2026. It is on by default through the Malicious URL protection setting (Teams admin center → Settings and policies → Messaging Safety); no separate configuration is needed.

Awareness points for staff (helpdesk and everyone else):

• Our IT team will never ask you to start a remote session from an unexpected Teams chat or call.
• Look for the "External" label next to a name. Accept a chat request only if you know the sender.
• If you are unsure, contact IT through a channel you already know, such as the service desk number on the intranet.
• A missing warning does not mean a message is safe; the new feature is an extra layer, not a guarantee.

If your organisation already blocks all external Teams chat, the feature changes little for you.

The expert view

Conceptually, this is the move from indicator-based to intent-based detection. Classic message security asks "is this URL or file known to be bad?" Social engineering in chat platforms usually carries no such indicator: the payload is the conversation itself, and the malicious step (approving a remote-control request, reading out a Quick Assist code) happens outside the message. A model that reads the exchange and asks whether the request fits the relationship is the logical answer, and email security vendors have been pitching the same shift for about a year as generative AI erases the spelling mistakes and clumsy phrasing that filters and users used to rely on.

What is genuinely new is scope rather than technique. Teams already showed spam or phishing warnings on some external chat requests, and Microsoft had added link protection, QR code blurring for external senders and other controls in 2026. This is the first time Microsoft has said it will apply language-model-style context analysis to Teams message content by default, across most tenants, with no opt-in.

Why it matters: helpdesk impersonation over Teams has become a reliable initial-access path for intrusion and ransomware operators, as Microsoft's September report shows. Attackers can also use AI to write fluent, tailored first messages at scale. A detection layer placed at first contact, before any remote session begins, targets the cheapest point in that chain to break.

The limits are also clear. The focus is first-contact external chats, so an attacker who has already compromised a partner's account, or who moves to a voice call, sits partly outside the stated scope. Microsoft has not said which model is used, how it performs, what the false-positive rate is, whether admins will be able to tune or disable it, or which licences include it. Warnings also depend on users acting on them, and alert fatigue is real.

The privacy trade-off deserves an explicit decision rather than a shrug. Microsoft's statements on purpose limitation and non-retention are clear, but they are commitments in a service notice, not technical controls customers can inspect. Regulated organisations should map them against their own data-processing records before December.

Still unknown, as of November 5, 2026: licensing, admin controls, the exact warning experience, and whether the detection will feed Microsoft Defender alerts for security teams. Microsoft has not revised MC1519866 since it was posted on October 21.

Official sources

• Microsoft 365 Message Center MC1519866 — "Microsoft Teams: Phishing detection for messaging" (October 21, 2026), via the mc.merill.net mirror
• Microsoft 365 Roadmap 575288 — "Microsoft Teams: Phishing Detection in Teams Messaging" (via mirror)
• Microsoft 365 Message Center MC1526651 — "Microsoft Teams: Malicious QR code detection in messages" (October 26, 2026), via mirror
• Microsoft Security Blog — "Impersonating IT support: Threat actors turn remote session into enterprise-wide access" (September 2, 2026)
• Microsoft Learn — Manage external meetings and chat in Teams
• Microsoft Learn — End-user reporting of security concerns in Teams
• Microsoft Support — Prevent spam or phishing attempts from external chats in Teams
• Coverage: BleepingComputer — "Microsoft Teams to get AI-powered, context-aware phishing detection" (October 23, 2026)