Deepfake detection leaves the cloud: Halo model scans video calls via neural processing chips

Cyber fraud has moved from fake emails and naïve phishing links straight to visual meeting screens. In 2024, the global engineering firm “Arob” lost $25 million after one of its employees was duped by a group video call that included company executives using deep-fake technology. The breach was not an isolated incident; a 2025 report from the FBI’s Internet Crime Complaint Center estimated total losses from corporate email breaches and deep-fake-enabled financial fraud at roughly $2.9 billion. Those losses reveal a critical defensive gap: traditional forensic tools always began operating only after financial transfers were completed and damage was done.

In response to this shift in targeting firms, Scam.ai (also known as Reality Inc.) launched a real-time deep-fake detection model called “Halo”, developed in engineering collaboration with chip maker Qualcomm.The model runs directly on the user’s device to spot and flag synthetic faces during live streams, before any executive decision is made based on what is seen or heard in the call.The system aims to protect major virtual-meeting platforms such as Zoom, Microsoft Teams and Google Meet, handling each video frame as soon as it arrives rather than waiting for downstream processing.

The technical shift in Halo’s architecture lies in its complete decoupling from cloud servers. The model relies on the “Snapdragon” neural-processing unit, Qualcomm’s silicon chip dedicated to on-device AI computing, to process each video frame locally the moment the meeting starts. This architecture eliminates the need to send video streams or audio recordings to remote servers and wait for a response, ensuring that no audio or visual data leaves the user’s device for cloud analysis. Engineering and marketing teams at both companies collaborated to fine-tune the model specifically for instant inference on this hardware, with availability currently limited to Windows systems and announced plans to support additional operating platforms in the future.

For executives and IT teams in the Gulf, Egypt and the broader Arab region, this development reshuffles protection priorities in digital meeting rooms. Family offices, investment funds and large regional corporations manage strategic transactions and financial transfers that increasingly rely on encrypted visual communication.On-device inspection ends the ongoing clash between strict data-governance policies that forbid routing board-room discussions to third-party clouds and the need to secure calls against identity spoofing.Protection now occurs at the digital edge of the display before any official endorsement is signed.

However, this path places regional organizations in an executive reality dictated by personal-device infrastructure budgets; protection is now tied to hardware and the type of processor inside a laptop rather than to a cloud-managed software add-on. The model’s current limitation to Windows PCs equipped with Snapdragon processors means IT departments will have to align deep-fake defense plans with desktop refresh and replacement cycles, otherwise leadership teams that rely on traditional compute platforms lacking neural processing units remain vulnerable to direct financial fraud through closed-screen calls.