What happened
Scam.ai announced Halo, a deepfake detection model that runs directly on Qualcomm's Snapdragon NPU (on-device AI chip) to flag synthetic faces in real time during Zoom, Microsoft Teams, or Google Meet calls. The tool processes video frames locally without sending any video or audio to the cloud, and is available now for Windows.
Why it matters
Video-call impersonation fraud is a growing threat to businesses—engineering firm Arup lost $25 million in 2024 after an employee was deceived by a deepfaked video of executives, and the FBI's 2025 IC3 report found business email compromise and deepfake-enabled wire fraud accounted for an estimated $2.9 billion in losses. Halo detects and flags synthetic faces the moment they appear on a call, before any decision is made, rather than after damage is done.
What to watch
Halo is currently available for Windows, with support for additional platforms planned.
Summaries like this, in your inbox every morning.
Deepfake-enabled fraud is emerging as a material financial risk for enterprises. The 2024 Arup case—a $25 million loss triggered by a single deceived employee—demonstrates that even large, well-resourced organizations remain vulnerable to video impersonation attacks. The FBI's 2025 IC3 report's finding of $2.9 billion in estimated losses from business email compromise and deepfake-enabled wire fraud suggests the problem is systemic and growing. Halo addresses a critical gap: most existing deepfake detection tools operate after the fact (post-call analysis, forensic review), leaving the decision-making moment unprotected. By detecting and flagging synthetic faces in real time—before any action is taken on the call—Halo shifts the detection from a forensic tool into an active defense. The technical design (on-device processing via Snapdragon NPU, zero cloud transmission) also addresses privacy and latency concerns that could otherwise slow adoption in regulated industries.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
A Zenn contributor sorted AI use into three modes — AI as an assistant, AI running alongside people, and AI th…
A developer released agent-guardrails, a one-script Python standard-library plugin that guards Bash, Write, Ed…
A Qiita article compared automated red-teaming methods for LLM apps — GCG, PAIR, and TAP — using mock implemen…
TOPPAN Holdings and Keio University ran a quantum-computer and generative-AI proof-of-concept from February 1…
In a webinar hosted by French Mathematical Society president Isabelle Gallagher, Fields Medal winner Hugo Dumi…
Anthropic said on October 15 it has started giving METR's six staff company devices, office space and access t…
