Scam.ai, the company building state-of-the-art AI to detect deepfakes and AI-driven fraud, announced the launch of Halo, a real-time, on-device deepfake detection model for video meetings, developed in close collaboration with Qualcomm.

Video-call impersonation fraud is a fast-growing threat to businesses. In 2024, engineering firm Arup lost $25 million after an employee was deceived by deepfaked video of company executives on a conference call. According to the FBI’s 2025 IC3 report, business email compromise and deepfake-enabled wire fraud accounted for an estimated $2.9 billion in losses.

Halo runs entirely on the user’s device, securing live Zoom, Microsoft Teams, or Google Meet calls and flagging a synthetic face in real time, before any decision is made based on what’s seen or heard on the call. It does this by running its detection model directly on the Snapdragon NPU, Qualcomm’s dedicated on-device AI silicon, processing each video frame locally as the call happens rather than sending it to a server and waiting on a response. No video or audio is ever sent to the cloud for analysis.

Scam.ai worked closely with Qualcomm’s engineering and marketing teams throughout Halo’s development, optimizing the detection model specifically for real-time inference on Snapdragon hardware.

“We didn’t want to build another tool that tells you about a deepfake after the damage is already done,” said Simiao (Ben) Ren, co-founder and CEO of Scam.ai. “Working with Qualcomm let us run real-time detection directly on the Snapdragon NPU, so Halo can catch a synthetic face the moment it appears on a call, without the video ever leaving the device. That’s the difference between a warning and a report.”