FR
live

The FBI dismantles FaceStealer, the North Korean campaign that filmed victims through fake video meetings

On 14 November 2026 the U.S. announced it had disrupted FaceStealer, a North Korean social-engineering campaign that stole more than $400 million in cryptocurrency through rigged video meetings, deepfakes, and ClickFix links. Verify any meeting invitation through an independent channel and never run a script presented to you as a "fix".

An empty office chair facing a laptop webcam, a single amber recording light glowing on the lens.

14 November 2026. BleepingComputer reports that the U.S. government has neutralized FaceStealer, a North Korean social-engineering campaign behind the theft of more than $400 million in cryptocurrency. 18 months. How long the campaign ran before the U.S. seized its cloud infrastructure. 900. The number of secretly recorded webcam videos of tech workers. Why it matters: the attack no longer relies on a cold-linked phishing email but on a video meeting in which the person you see is a recording of someone you know — and that someone was already a victim.

The mechanism: the person on screen is not there

The core of FaceStealer is a simple, brutal inversion. The attacker does not merely impersonate an account: they hijack accounts on Telegram, LinkedIn, Slack, and Discord, then use them to invite the target to a video meeting under a plausible pretext — a code review, an interview, a payment approval.

While the target waits for “the meeting to start”, the attacker secretly records their webcam. That captured footage then becomes bait for the next victim: on the following call, the attacker plays back a recording of a familiar person — a colleague, a partner, an investor — who appears present and attentive in the stream. Some meetings combined recordings of several different people to raise the pressure. The victim believes they are talking to their network; in reality they are watching a tape of previous victims.

The pivot comes shortly after. The attacker claims a “technical issue” and displays a fake error message, then offers a “fix” — an installer or update to download. This is the ClickFix pattern, the most widespread phishing technique of the year: the link installs malware that steals crypto data, credentials for video platforms (Zoom, Google Meet, Microsoft Teams), contact lists, social media, and webcam footage.

An arsenal that industrialized face fraud

What separates FaceStealer from earlier campaigns is the assembly of three components that had never been combined at this scale.

The first is deepfake by reuse. The attacker does not need to synthesize a face: they replay real recordings stolen from earlier victims. The quality is perfect because the video is real — only the identity of the person on screen is false. The second is ClickFix, which turns a staged failure into an infection vector: the victim, embarrassed by a “problem” in a professional meeting, executes the payload themselves. The third is systematic webcam collection, which feeds the campaign with material for subsequent attacks — a self-sustaining loop.

Liveness checks, the standard countermeasure against video fraud, fail here for a precise reason: the footage is genuine. Blink detection, head-turn prompts, and challenge-response checks exist to stop synthesized or pre-recorded video, but FaceStealer replays real recordings of real people — captured minutes or days earlier from a previous victim — so the stream passes every software test. What those tests cannot verify is the one thing that matters: who is actually on the other end. That is why the FBI’s defensive guidance converges not on a technical control but on a behavioral one: verify the invitation out of band, join only through the official app, and leave the webcam off. The camera is not the victim here — it is the weapon the attackers use to build their next lure.

ArcticWolf identified 677 targets across 58 countries, with a focus on founders and senior executives in the crypto and Web3 sectors. The researcher also documented additional lures: fake financial documents, fake PDF reader updates, identity-verification prompts, and fake security audits — all designed to push the target into running the “fix”.

Patience as a weapon: waiting for the wallet to grow

The most chilling detail in the FBI file is temporal. According to the federal bulletin, attackers sometimes delayed targeting by months, or more than a year, after the initial compromise. The reason is economic: they waited for the victim’s cryptocurrency holdings to grow before striking.

That delay changes how you read the risk. A silent compromise is not a closed incident: it is a dormant position that wakes up when the value at stake reaches a threshold. For a victim, it means the infection may have preceded the theft by many months — and that the entry point (the fake meeting) and the exit point (the wallet drain) are uncorrelated in time. Retrospective detection must therefore reach back far further than the date of the theft.

The expanded target: open-source maintainers

The FBI bulletin documents a worrying extension of the campaign into the software supply chain. In one incident it describes, the attackers compromised the maintainer of a popular JavaScript library, downloaded tens of millions of times a week, after luring them into a fake Teams meeting with several participants — including one face familiar to the developer.

That attack was thwarted by fast detection, but the FBI confirms multiple other attempts against developers in the npm and Rust crates ecosystems. The reasoning is clear: compromising a maintainer distributes malicious code to millions of dependents in a single operation, without targeting each end user. Amazon had already linked supply-chain attacks on npm — including Debug Chalk — to BlueNoroff in a July 2026 report. FaceStealer confirms that the maintainer route has become a first-rank target for Pyongyang.

Attribution: BlueNoroff, Lazarus’s crypto arm

ArcticWolf attributes FaceStealer to BlueNoroff, the Lazarus subgroup that specializes in cryptocurrency theft and attacks on the financial sector. Huntress had already documented in 2025 a similar campaign in which North Korean attackers used Zoom calls with deepfaked executives to infect leaders with custom macOS malware.

The attribution matters for defense. BlueNoroff is a funded, disciplined unit whose operations feed North Korea’s weapons program. It treats crypto theft as a production line: standardized lures, calculated patience, and systematic reuse of stolen material. The $400 million from FaceStealer adds to the $6.36 billion Binance heist — on which the Department of Justice separately announced it had frozen more than $2.7 billion and obtained court authorization to seize more than $1.2 billion.

The verdict

If you operate in crypto, Web3, or open source, treat the identity on screen as unverifiable from now on: confirm any meeting invitation through an independent channel (a signed message, an outbound call to a known number), join only through the official app with an invite code, and keep the webcam off unless explicitly required.

If someone presents you a “fix” or “installer” during a meeting, never run it — no legitimate technical problem is solved by a script received from a remote party. And if you maintain a popular package on npm or crates.io, enable hardware multi-factor authentication, watch for anomalous publications, and distrust improvised meetings with “partners” you have never met in person. FaceStealer has been dismantled, but the technique — a trusted person’s face replayed on tape — belongs to no one and will be reused.

References

The cyber brief, every Tuesday

The flaws that matter and the patches to apply, in a ten-minute read.

No spam. One-click unsubscribe.
read next

On the same topic

A hacker hijacks Optimizely scripts and serves a card skimmer to CNN, BBC, and Zoom

On November 17, 2026, Optimizely confirmed that a flaw in its A/B testing product let an attacker inject malicious code into scripts loaded by 178 organizations, including CNN, the BBC, and Zoom. Inventory the third-party scripts your pages run and lock down your vendors’ custom JavaScript.

UnrealIRCd 6.2.7.1 closes an exploited flaw that hands server authority to any user

CVE-2026-114713, rated 8.6, lets any user of an IRC network with linked servers inject server-to-server commands through the label tag of the labeled-response module. Added to CISA’s KEV catalog on 13 November 2026 with a 16 November deadline, it is already being exploited: apply the hot-patch without restarting, or blacklist the module.

← Back to the feed

Type at least two characters.

↑ ↓ navigate ↵ open esc dismiss