Is the Person on Your Video Call Even Real?

Deepfake fraud in hiring is up over 1,000% in a year. Scam.ai and Qualcomm just built a chip-level answer.

Welcome Back,

At Computex 2026 in Taipei, deepfake detection startup Scam.ai announced a partnership with Qualcomm and launched Halo, an on-device deepfake detection model built for live video calls, according to the company's own announcement. Both pieces of news landed at Qualcomm's booth as part of the event's Agentic AI track.

The pitch is straightforward, Halo runs quietly in the background of any video call, checking whether the face on your screen is a real, unedited human or an AI-generated synthetic feed, without sending your video anywhere. It's aimed squarely at two groups, HR teams running remote interviews, and executives who take a lot of high-stakes calls.

Today we look at what Halo actually does, why on-device processing is the detail worth paying attention to, and how the deepfake hiring fraud numbers Scam.ai cites hold up against independent research.

🛡️ What Halo Actually Does

Halo operates in the background of any video conferencing session and flags synthetic or AI-generated video in real time, according to Scam.ai's announcement. The key word there is real time, this isn't a tool that reviews a recording after the fact, it's meant to catch a fake feed while the call is actually happening.

The company frames the threat in blunt terms. "Deepfakes has rapidly risen as one of the major concerns for enterprises in multiple fronts, all traditional security measures are oblivious once a human is breached," said Dennis Ng, co-founder of Scam.ai.

That framing, attacks getting curbed "from the source," is really about where trust breaks down in a video call. Most enterprise security stops at the network or the login screen. Once a person appears on camera and starts talking, older security tools have nothing left to check, they simply assume the human in front of the camera is who they claim to be.

Halo is optimized to run directly on Qualcomm-powered hardware rather than routing video through the cloud.

🔒 Why On-Device Processing Is the Real Story

The Qualcomm partnership gives Scam.ai access to device ecosystem resources and optimization support, enabling Halo to run locally on personal computers without relying on cloud infrastructure, according to the announcement. No video footage leaves the user's computer at all.

That local-first approach also means Halo doesn't add a network round-trip that could introduce lag during a live call, and it works whether or not the connection to the internet is fast or stable, both practical advantages for something meant to run passively in the background without anyone noticing.

🎯 Who Halo Is Actually Built For

Scam.ai names two specific target audiences, HR and recruiting teams conducting video interviews, and high-value executives, including CEOs, CFOs, and venture capitalists, who take frequent, high-stakes calls. Those are genuinely different risk profiles that happen to converge on the same underlying vulnerability, trusting the face on the other end of a video call.

Halo is described as passive, running in the background with no change to existing call workflows, meaning it doesn't require anyone to install a separate app or take an extra step before joining a meeting, at least according to Scam.ai's own materials. Enterprise integration details and additional platform partnerships are expected in the coming months.

HR teams running remote interviews are one of Scam.ai's two named target audiences for Halo.

📊 Do the Deepfake Hiring Numbers Hold Up?

Scam.ai's press release cites a specific figure, deepfake fraud attempts increasing over 2,000% in the past three years, and it's worth checking that against independent research rather than taking a vendor's own statistic at face value.

The independent data broadly supports genuine, rapid growth, even if the exact multiplier varies by source. Pindrop's 2025 Voice Intelligence Report found deepfake attempts in hiring jumped 1,300% year over year. Separately, an analysis of 19,368 live interviews found 38.5% of candidates were flagged for AI-cheating behavior between July 2025 and January 2026, a rate that tripled from 9% to 45% in just three months. A 2025 Checkr survey found 41% of IT, cybersecurity, risk, and fraud leaders confirmed their organization had unknowingly hired a fraudulent candidate.

That last figure is arguably the single most important number in this whole story. Humans genuinely cannot reliably tell real video from synthetic video anymore. Whatever the exact percentage Scam.ai cites, the underlying trend, rapid growth in deepfake fraud attempts paired with human detection collapsing toward chance, is well-documented across multiple independent sources.

Worth flagging directly: this is a company's own press release, and Scam.ai has an obvious commercial interest in emphasizing the scale of the problem its product solves. The independent numbers back up the general trend, but the specific 2,000%+ figure comes from Scam.ai itself and hasn't been independently verified against a named methodology.

⚖️ Detection Is an Arms Race, Not a Fix

It's worth being honest about what a tool like Halo can and can't promise. Detection tools and generation tools tend to improve in tandem, each new detection method eventually prompts a new generation technique built to slip past it.

None of that erases the genuine value of building detection directly into the device layer. It just means the honest framing is "a meaningful new layer of defense," not "the problem is solved," a distinction that matters given how fast the underlying fraud numbers are still climbing across every independent source that's tracked them.

High-stakes executive calls are the other named use case Scam.ai is targeting with Halo.

🧠 AI Spotlight Analysis

The genuinely interesting part of this story isn't the specific product, it's what it represents, deepfake defense moving from something that happens after the fact, forensic analysis of a suspicious recording, to something that happens live, at the exact moment a fake video is being used to deceive someone.

The Qualcomm partnership matters for a similar reason. Building detection into the chip layer, rather than as an app-level add-on, is the kind of infrastructure decision that could make real-time deepfake checking a default feature of business laptops, rather than something IT departments have to separately deploy and maintain.

The honest read: with humans now detecting deepfakes barely better than a coin flip, tools like Halo aren't optional extras anymore, they're becoming a genuinely necessary layer of defense. Whether Halo specifically holds up under independent scrutiny is still an open question, but the underlying need it's responding to is very real and well-documented.

🔗 Sources and Further Reading

Thanks for reading AI Spotlight.

Our mission is simple: deliver clear, trustworthy, and actionable AI insights that help professionals stay ahead without the hype.