Microsoft Plans AI Phishing Detection in Teams

Microsoft says Teams will begin using AI to spot phishing inside chats from early November, a defensive feature aimed at messages from external contacts who reach a user for the first time. The plan was reported by BleepingComputer on 23 October, citing a Microsoft 365 Message Center update posted on Wednesday, and Microsoft’s own support documentation now describes the capability.

What is new here is the target. Rather than scanning only for malicious links or attachments, the system reads the context of a conversation to judge intent. “Microsoft Teams is introducing AI-powered, context-aware phishing detection to help identify phishing, social engineering, impersonation attempts, and deceptive intent,” the company said in the Message Center notice, adding that it “analyzes conversation context to identify suspicious behavior that may not include traditional indicators such as malicious links or files.”

In brief

Microsoft says Teams will start using AI, context-aware phishing detection on external first-contact chats, with a targeted release planned for early-to-mid November.
Worldwide general availability is planned for December, which Microsoft describes as “likely” rather than fixed; no admin action is required.
The company says message content is not retained for AI training or other secondary uses, and the feature is for security only.

How the detection is meant to work

By Microsoft’s account, the check runs quietly at first. During an initial quality-validation phase it may operate in the background without showing anything; once validated, a user may see a warning when a message looks like phishing, giving them a moment to pause before replying. For most users the focus is external first-contact chats, the higher-risk scenario in which an unknown sender opens a conversation. Teams mobile users in China may get additional protection that also covers in-tenant conversations, aimed at fraud patterns seen in that region. No admin configuration is required to switch the protection on, and Microsoft asks organisations to remind people to report both suspicious activity and any detection they think is wrong.

A defensive use of AI, with the usual caveats

This is AI pointed at attackers rather than at productivity, which is why it reads as a genuine tech-for-good move. It also fits a widening Teams security push: Microsoft has lately moved to let admins lock external users through the Defender portal, to blur QR codes from external senders, and to support deepfake and impersonation checks in meetings. The logic mirrors other defenders turning detection on scams, from Chrome’s expanded “suspicious site” warnings to Microsoft’s own effort to treat AI agents as insider risks and watch them with guardian agents. The threat it targets is real and growing, as criminal crews increasingly lean on conversation-based deception; researchers have even shown attackers wiring up Claude and Grok agents to hit banks. The caveats are the familiar ones for any model that judges intent. Flagging “deceptive intent” without a malicious link means some legitimate messages will trip the warning, which is why Microsoft includes a path to mark a detection as not a concern. And the dates are a plan, not a shipped product: the targeted release is pencilled in for early-to-mid November and the worldwide rollout only “likely” for December.

What to watch

The questions that matter are accuracy and reach. Teams is a favourite entry point for social-engineering crews, including ransomware gangs posing as help-desk staff, so a warning that appears at the right moment could blunt a common opening move, if it fires accurately and does not train users to click through alerts. Worth tracking: whether the November target holds, how Microsoft handles false positives once warnings go live, and whether the China-only in-tenant coverage expands elsewhere. For now it is a promised layer of defence, built into a tool that attackers already know how to abuse.