New Teams Feature Detects Synthetic Media

Microsoft is introducing a new feature in Microsoft Teams that will alert participants if someone in a meeting is using a cloned voice or an AI-generated face. The detection of synthetic media will be handled by third-party providers, which companies must select and authorize to analyze their meetings.

Monitoring System at Meeting Entry

A monitor will be placed at the entrance of meetings to manage these checks. The first companies to receive the feature will be those enrolled in the Targeted Release program, expected to get it between mid-November and the end of November on Teams for Windows, Mac, and web browsers.

Setup Process for Companies

According to Microsoft's documentation, the detection of synthetic media is performed by third-party providers, and the display of alerts is reserved for Teams. Companies will need to allow an external bot into their video conferences, two months after receiving from Microsoft the means to automatically block external bots.

Alert Details and Reporting Rules

Each report is accompanied by a confidence score between 0 and 1 and a severity level on three levels. Microsoft cites voice cloning and substituted faces as examples and sends the alert to all participants, including the organizer. Editors should not report a person more than once per minute.

Complement to Existing Reporting Tools

With this automatic signal, Microsoft complements the "Report a concern" button, scheduled for this month, which allows a participant to report identity theft or phishing attempts. However, it is necessary for a human to detect the fake.

Mobile Users Lack the Feature

Microsoft's note does not mention iPhone and Android applications, and on your phone, this button will therefore be your only recourse for now. Microsoft has not yet named any certified editors, nor specified the necessary licenses, notes Windows Latest.

Privacy Concerns with Third-Party Bots

When Teams improves the sound and image of your meetings with Microsoft's AI, the company promises that the audio and video "never leave Teams and are not stored anywhere by anyone." Microsoft does not make this commitment for an external editor's bot.

Third-Party Detection Tools

Sensity markets a detection bot compatible with Teams and states in its documentation that it joins the call and captures 30 seconds of audio and video at each analysis. Microsoft does not say if it will certify Sensity, nor how long these editors will keep your conversations, nor if they can use them to train their models.

Real-World Risks of Deepfakes

Companies have good reasons to try the experience. In 2024, an employee of Arup in Hong Kong transferred 25 million dollars to scammers after a video call where his financial director and colleagues were all deepfakes. However, with a score of 0.8, you will only have a probability, and Microsoft has not published any error rate of these detectors.

Human Verification Still Needed

Abhijith M B, a journalist at Windows Latest, doubts their reliability with a mediocre webcam or a choppy connection. Before approving a transfer requested during a video call, therefore, call your interlocutor on a number you know, with or without an alert.

Future Microsoft Detector Mentioned

In the PowerShell documentation of Teams, Windows Latest has identified the mention of a Microsoft-signed detector reserved for a future version. The publisher has not confirmed this project.