Home » Deepfake Hyperrealism and FaceStealer Scheme Expose Video Call Verification Failures

Deepfake Hyperrealism and FaceStealer Scheme Expose Video Call Verification Failures

By

–

Deepfake Hyperrealism and FaceStealer Scheme Expose Video Call Verification Failures

The FBI has seized the cloud infrastructure behind a North Korean scheme that stole more than $400 million in cryptocurrency over 18 months, a campaign the bureau calls FaceStealer. The operation relied on secretly recorded webcam footage of earlier victims, replayed in fake Zoom and Teams calls to trick new targets into installing malware.

The Technical Reality of FaceStealer

The scheme does not depend on deepfakes in the way most people imagine them. Instead, attackers hijack a victim’s Telegram, LinkedIn, Slack or Discord account and message that person’s contacts to propose a meeting. When the target joins the call, they see unaltered footage of someone they know sitting silently. The attackers claim in the chat that the target’s audio is not working, so the silence seems like a technical glitch. The target is then coaxed into installing malware that records their own webcam, feeding the next iteration of the cycle.

The FBI alert describes the infrastructure as containing more than 900 of these recordings. According to the Justice Department, FaceStealer also deploys deepfake videos of well-known figures in the cryptocurrency and Web3 sectors to make the approach more convincing. The bureau’s Cyber Branch Operations Director, Brett Leatherman, confirmed that North Korea uses proceeds from large-scale cryptocurrency theft to fund its weapons programmes.

Why Video Verification Is Failing

The Office of the Comptroller of the Currency lists video verification among the controls banks use against impersonation fraud. Both FBI alerts describe North Korea faking who appears on camera, undermining that control. A bank employee who joins a fake call on a work computer could give attackers access to the employer’s network, the alert notes. The technical reality is that the face on screen is often a real person, just not the person the target thinks they are talking to.

man in blue dress shirt smiling beside woman in black and red floral dress
Photo: visuals / Unsplash (Unsplash License)

Deepfake Hyperrealism and the Brain

New research from the University of Florida, published in October in Social Cognitive and Affective Neuroscience, helps explain why synthetic faces work so well. The study found that AI-generated faces appear more realistic than photographs of actual people, a phenomenon the researchers term “deepfake hyperrealism.”

Under the hood, the bias links to reduced activity in the anterior insula, a brain region that acts as a hub for detecting salience and information that warrants caution. When participants viewed deepfakes, that region quieted down. Both younger and older adults rated AI faces as more real than real photos, but older adults showed a significantly stronger hyperrealism bias, making them more vulnerable to scams involving fake online identities.

The Scale of the Problem

Deepfake incidents have surged. Surfshark research analysing the AI Incident Database and OECD data identified 2,430 incidents between 2020 and the third quarter of 2026, with nearly 90 per cent occurring from 2024 onwards. The count hit 970 in 2025, and 809 were documented in the first three quarters of 2026 across 114 countries. Deepfake-enabled fraud has cost an estimated $3.91 billion globally since 2020.

The technology leverages generative adversarial networks and variational autoencoders to create synthetic media depicting real or fictional people. Non-consensual sexual deepfakes now affect ordinary people through accessible “nudify” apps and AI generators; a survey of 7,230 adults in Australia, the UK and the US found 7 per cent reported being victims.

Practical Detection

The FBI’s guidance is blunt: end the call if something feels off. Verify the person’s identity through a separate, trusted channel. Be wary of any meeting where the other side claims audio problems but video works perfectly. Check for the malware installation prompt that follows — the detail that matters is the request to run code or download a file to “fix” the connection. The infrastructure seizure disrupts the current operation, but the method is portable and the hyperrealism bias is human.

Sources

Photo: Chris Montgomery / Unsplash (Unsplash License)

Get breaking stories first on our Telegram channel.

More stories by

Most read

Crypto

Internet & Data Centers

Cybersecurity

Business