Microsoft’s current Teams guidance describes AI-based phishing detection that evaluates conversation context in one-to-one and group chats with new external contacts. It covers mobile, desktop and web; mobile users in China may receive additional checks within their organization.
The guidance says users can block and report a confirmed phishing contact or mark a mistaken warning as not a concern. Microsoft says the detection is used for security, rather than model training or unrelated purposes. The support page does not establish a dated rollout schedule or confirm activation in any particular organization.
NAOAT analysis
For an employee, the practical question remains whether the request makes sense. A familiar company name or a message that contains no link should not settle that question. A person claiming to be technical support can still ask for an action that deserves independent verification. Training should therefore explain where employees can find their organization’s genuine help desk, rather than asking them to negotiate identity inside a suspicious conversation.
A warning is most useful when the next step is simple. Organizations could provide a short internal procedure covering how to stop the exchange, preserve enough information for review and contact the responsible security team. That procedure should avoid requiring employees to forward sensitive material into additional chats. It should also explain how legitimate external contacts can continue their work after a mistaken alert is resolved.
For administrators evaluating the feature, alert volume alone would be an incomplete measure of value. More useful questions include whether employees understand warnings, whether reports reach someone who can act and whether repeated mistaken detections interfere with normal collaboration. The goal is a dependable response around the detection system. No warning should be interpreted as a guarantee that an unfamiliar request is safe, and a warning should begin a review rather than automatically establish wrongdoing.