Halo by Scam AI

Product Introduction

1. Definition: Halo by Scam AI is a real-time, on-device deepfake detection software for video conferencing. It is a desktop application that uses local machine learning models to analyze video streams for synthetic media, specifically targeting AI-generated faces and real-time face-swaps during live calls.
2. Core Value Proposition: It exists to prevent financial fraud and security breaches by detecting deepfake video calls in real time. Its primary value is enabling users to verify the authenticity of participants on platforms like Zoom, Microsoft Teams, and Google Meet before authorizing transactions or sharing sensitive information, directly addressing the emerging threat of synthetic-executive fraud.

Main Features

1. Real-Time, On-Device Analysis: Halo performs frame-by-frame analysis of video conferencing feeds directly on the user's Windows PC, with no data uploaded to the cloud. It processes video roughly four times per second, using locally stored neural networks to detect artifacts and inconsistencies indicative of AI-generated or swapped faces. This happens entirely in system memory.
2. Cross-Platform Tray Application: The software installs as a standalone Windows application that runs in the system tray. It requires no plugins, browser extensions, or integration with the conferencing apps themselves (Zoom, Teams, Meet, WebEx, Slack). It passively captures and analyzes the video output from these applications.
3. In-Call Alerting & On-Device Audit Trail: When a synthetic face is detected, Halo surfaces a visual alert during the live video call, allowing for immediate intervention. All analysis logs and detection events are stored locally on the device, creating a secure audit trail for post-incident fraud investigation and compliance review without ever exposing call data externally.

Problems Solved

1. Pain Point: The inability to authenticate participant identity during routine video calls, leading to business email compromise (BEC) and wire transfer fraud via real-time deepfake impersonation. Traditional security (email filters, 2FA) is bypassed as the attack occurs within the trusted context of a live conversation.
2. Target Audience: Finance and treasury teams (CFOs, AP staff, controllers), executive assistants, HR and recruiting professionals conducting remote interviews, legal teams handling sensitive discussions, and any business personnel authorized to make high-value approvals based on video call verification.
3. Use Cases: Stopping a fraudulent wire transfer by flagging a deepfake CFO on an approval call; preventing a social engineering attack during a simulated "IT support" call; verifying the identity of a remote job candidate during a final interview; providing an audit record for compliance after a sensitive executive briefing.

Unique Advantages

1. Differentiation: Unlike cloud-based detection APIs, Halo operates with near-zero latency (no network round-trip), maintains complete data privacy (no video upload), and incurs no per-minute GPU costs. Unlike platform-native features (e.g., Microsoft's "Speaker Coach"), it is specifically engineered to detect synthetic media, not just liveness or presentation skills, and works agnostically across all major conferencing platforms.
2. Key Innovation: Its fully localized detection architecture. By keeping the entire detection pipeline—from video capture to ML inference to alerting—on the endpoint, it eliminates the cloud attack surface, guarantees privacy by design (frames are analyzed in memory and discarded), and enables truly real-time protection synchronized with the flow of a live conversation.

Frequently Asked Questions (FAQ)

1. How does Halo by Scam AI detect deepfakes on video calls? Halo uses on-device machine learning models trained to identify subtle artifacts, inconsistencies in lighting, texture, and physiological signals (like unnatural blinking or facial micro-movements) that are characteristic of AI-generated or face-swapped video, analyzing the call feed locally in real time.
2. Is Halo compatible with Google Meet and Microsoft Teams? Yes, Halo is a standalone Windows application that works across major video conferencing platforms including Zoom, Microsoft Teams, Google Meet, WebEx, and Slack Huddles without requiring individual plugins or extensions for each service.
3. Does Halo record or store my video calls? No. Halo is designed with privacy-by-design principles. Video frames are processed in your computer's RAM (system memory) for analysis and are immediately discarded after a detection verdict is made. Nothing is permanently recorded, stored on disk, or transmitted over the network.
4. What is the difference between Halo and cloud-based deepfake detection? The core difference is latency and data privacy. Cloud detection requires uploading video frames, leading to delays that are impractical for live call intervention. Halo's on-device processing provides instant alerts and ensures your sensitive call data never leaves your computer, reducing exposure to data breaches.
5. Can Halo prevent the type of deepfake scam that happened to Arup? Yes, Halo is specifically engineered to address "synthetic-executive wire fraud" as demonstrated in the 2024 Arup case (~$25M loss). By flagging synthetic faces in real time during the approval call, it aims to give employees a critical verification tool before authorizing transactions, directly mitigating this specific, high-value attack vector.