Back to articles

FBI disrupts FaceStealer, North Korea's fake video-call scheme tied to $400 million in crypto theft

What happened

On November 12, 2026, the US Department of Justice and the FBI announced that they had seized cloud-based infrastructure used in a North Korean social engineering scheme the FBI calls FaceStealer, which relies on imposter versions of Zoom, Microsoft Teams and Google Meet (DOJ). According to the department, the scheme has resulted in the theft of more than $400 million in cryptocurrency over the past 18 months. Under a federal warrant, the FBI seized more than 900 webcam videos of victims that the operators had recorded in secret and reused to impersonate those victims in later attacks.

The FBI published a FLASH advisory the same day, numbered AC-111226-MS and marked TLP:CLEAR, aimed at the cryptocurrency and web3 industry and at open-source software maintainers (FBI FLASH). It attributes the activity to North Korean state-sponsored hackers tracked under several names, including Sapphire Sleet, UNC1069, Dangerous Password, CryptoCore, APT38, BlueNoroff and Stardust Chollima. The bureau also published a summary page with a short version of the attack cycle (FBI).

The disruption was led by the US Attorney's Office for the Central District of California and the FBI's Los Angeles Field Office. The same announcement covered a separate action: the court-authorized seizure or restraint of more than $1.2 billion of the cryptocurrency stolen from Binance on or about October 29. The department attributes that theft to a different North Korean group it calls TraderTraitor and presents it as separate from FaceStealer (DOJ). Reuters also described the FaceStealer cluster as distinct from the hackers linked to the Binance theft (Reuters via Investing.com). This article covers FaceStealer only.

How the scheme works, according to the FBI

The FBI describes a six-stage cycle in which each victim supplies the material used to deceive the next one (FBI FLASH).

It starts with a person whose device is already compromised. The operators scan that device for cryptocurrency wallets and seed phrases, take over the person's Telegram account or another encrypted messaging app, read their contact lists and chat history, and secretly record their webcam while they wait for a video call to start.

Using the hijacked account, the operators then message the victim's contacts and propose a Zoom or Teams meeting to catch up, discuss a collaboration or talk about an investment. Because they can read past conversations, they can tailor the invitation to each contact. The FBI says the hackers have also created new accounts, and hijacked established ones, on Telegram, LinkedIn, Slack and Discord to make first contact.

When a target agrees, the operators send a Calendly or similar scheduling invite. The FBI notes that these scheduling emails often contain no meeting link. The link arrives shortly before the meeting, sent from the hijacked account, and points to one of many typo-squatted domains the operators registered to resemble Zoom and Teams. The advisory gives two examples, uu01webzoom[.]us and teams.livemeet[.]us. The page behind the link is a copy of the Zoom or Teams interface that the FBI describes as designed to be indistinguishable from the real one.

Once the target is admitted, the operators play the recorded webcam video of the person who invited them, often alongside other recordings so the call appears to have several participants. Acting as the trusted contact, they send a message saying the target cannot be heard and guide them to click an update prompt, run a script or paste code into a terminal. Following those instructions installs a browser cookie harvester, keyloggers, screen capture tools and a backdoor. The target's own webcam has been recording since they clicked the link, so they become the next source of video, contacts and access.

The FBI adds two points that widen the risk. First, the operators have begun using what the advisory calls super intelligence, or SI, to turn public photos and videos of well-known people in the cryptocurrency and web3 industries into deepfakes that can appear as call participants. The bureau says the scheme still relies overwhelmingly on genuine stolen video. Second, they have used the same technique against open-source maintainers. The advisory describes a maintainer of a JavaScript library with tens of millions of weekly downloads who was drawn into a fake Teams call earlier this year, shown at least one participant he recognized and infected. The operators then used his access to create malicious dependencies that would be downloaded along with the library. The FBI calls it a failed attack and credits quick detection and response. It does not name the library, and it reports further attempts against maintainers on npm and maintainers of Rust crates.

Thefts did not always follow quickly. In several reported cases, the FBI says, the operators waited months to more than a year after the initial compromise while wallet balances grew. The advisory also notes that an infected device can give the operators access beyond the victim's own wallets, such as an employer's network or a maintainer's software repository account.

What Arctic Wolf adds

The FBI advisory credits Arctic Wolf, Jumpsec and Microsoft Threat Intelligence for earlier technical reports. Arctic Wolf Labs published an update on the same day as the announcement, based on what it describes as continuing access to the operators' media library (Arctic Wolf). These are a vendor's findings, not government statements.

Arctic Wolf reports that the number of people for whom the actor had prepared material grew from roughly 400 in its April report to 677 people across 58 countries, and that founders and C-suite executives make up 84 percent of them. The company stresses that these counts describe who was prepared against, not confirmed compromises. It says the operators now screen-record YouTube interviews and conference talks to capture a specific person's likeness and voice, which lets them build fake meeting footage of someone they have never compromised. It also reports that the actor now combines several generative AI tools, that its forensic review flagged 349 videos in the library as likely modified as of November 6, and that it based deepfake determinations on provenance evidence because the output could no longer be told apart from genuine webcam footage by eye.

Arctic Wolf also says the fake meeting is no longer the only lure. It describes new pretexts built around shared financial documents, a fake Chrome PDF viewer update, a fabricated media player called NovaStream, identity verification prompts, hardware wallet access and security audits. Each follows the same pattern: build professional trust, manufacture a technical problem and persuade the target to run a "fix". The company reports 98 high-confidence domains in the current infrastructure, 73 of them not previously reported.

Microsoft's April 2026 Sapphire Sleet analysis, also cited by the FBI, describes an earlier delivery method built around a fake Zoom SDK update file (Microsoft).

What is confirmed and what is not

The confirmed elements come from the DOJ release and the FBI advisory: the seizure of cloud-based infrastructure, more than 900 seized victim videos, the figure of more than $400 million over the past 18 months, the attribution and the stage-by-stage method. These are government statements. The release does not describe criminal charges in the FaceStealer part of the action.

Several details remain unclear. The 900 figure has been framed in different ways. The DOJ release and the advisory's text describe more than 900 seized victim webcam videos, the advisory's title refers to a webcam library of more than 900 tech workers, and the BBC reported "more than 900 victims" (BBC). The FBI has not said how many of the people recorded lost funds, so the figure is best read as the size of the seized video library, not a count of theft victims. The government has not named the people or companies that lost funds, the open-source library or the full set of fake meeting domains, and it has not said how much of the operators' infrastructure remains active after the seizure. Arctic Wolf's descriptions of new lures rest on its own access and analysis.

We checked the two example domains from the FBI advisory through public registration data and DNS on November 13, 2026, without visiting them. Registry RDAP lists uu01webzoom[.]us as registered on January 23, 2026 through NameSilo, with no hold status, and the name did not resolve to an address during our check (RDAP record). livemeet[.]us carries a 2011 creation date and a last change in March 2025 at GoDaddy, and the teams subdomain named by the FBI returned no DNS record (RDAP record). These records do not show when the operators obtained either name or when they stopped using it.

Why this matters to enterprises

FaceStealer is mainly a crypto theft scheme, but the method travels. It abuses three things most organizations rely on: the identity of colleagues and partners, the familiar look of mainstream meeting tools and the public likeness of senior leaders. The FBI's account shows how one compromised person can hand an attacker a working identity, a contact list and a video clip that make the next approach look like an ordinary call with a known contact. For companies whose staff manage treasury, signing keys, code repositories or investor relationships, a single personal compromise can become a route into the business and its partners. Arctic Wolf's finding that most prepared targets are founders and executives puts leadership exposure squarely in scope.

Where Outtake could have helped

The analysis below is hypothetical. It does not suggest that any victim used or lacked Outtake, or that any part of FaceStealer would have been prevented. Much of the scheme happens in private chats and on the target's own computer, where an external service has no view. Several stages, however, depend on public assets: lookalike domains, new accounts on professional platforms and synthetic video of recognizable people. Earlier identification could matter there.

Spotting meeting-themed lookalike domains

FaceStealer's link step depends on domains built to pass a quick glance as Zoom or Teams. Outtake lists phishing and lookalike domains as a core brand protection use case and says its agents read logo, layout, image, video and audio, not only text (Outtake brand protection). For a targeted organization, the more practical value lies in investigation. Outtake says that from a single flagged domain its agents trace shared hosting, registration patterns, analytics IDs, linked social accounts and operator activity, and can expose pre-staged infrastructure that has not yet been activated. If an employee reports a suspicious meeting link, that tracing could help a security team find sibling domains and block them before other staff receive them.

The limit is plain. These domains imitate meeting platforms, not the targeted company, so monitoring a company's own brand names would not find them by itself. Coverage would depend on deliberately watching meeting-themed patterns or on a starting signal such as a reported link.

Finding fake profiles of executives and staff

The FBI says the operators created new Telegram, LinkedIn, Slack and Discord accounts as well as hijacking real ones. New accounts that use an executive's or employee's name and photo are impersonation in the ordinary sense. Outtake's executive protection page says it takes down impersonation accounts at the source and detects threats across text, images, video and audio (Outtake executive protection). Removing a fake LinkedIn or Telegram profile of a founder before it reaches investors or portfolio contacts could take away one opening move.

Hijacked accounts are different. When an attacker controls a person's genuine Telegram account, there is no fake profile to find. The account is real, its history is real and the messages arrive in private chats. Outtake has no view of those conversations unless the owner or a contact reports them, and recovery runs through each platform's own account-restoration process.

Watching for synthetic video of known leaders

The FBI and Arctic Wolf both describe deepfakes built from public footage of well-known people. Outtake's executive protection material names deepfakes and cloned voice among the threats it looks for on social platforms, fringe forums, the open web and the dark web. Where synthetic clips of a leader, or the public talks used to make them, are being circulated or repurposed in public, earlier detection could support warnings to staff and partners and takedown requests.

The footage FaceStealer plays during a call is shown inside an attacker-controlled meeting page, not posted publicly, so it may never appear anywhere an external service can find it. Arctic Wolf's note that its own deepfake findings rested on provenance evidence, not on how a video looked, is also a reminder that visual detection has limits.

Turning one report into a map of the operation

Outtake describes a Recon Agent that takes one signal and maps the operation behind it, with the example of linking an X account, a Telegram group, a WhatsApp channel and a payment link (Outtake executive protection). Its platform page lists evidence packaging, automated reporting and platform escalation as parts of the workflow (Outtake platform). Applied to a FaceStealer-style report, that could mean connecting a reported meeting link to related domains, a scheduling page and any public impersonation accounts, then packaging evidence for registrars, hosts and platforms. Any such outcome depends on those providers acting, and on the operators not having moved on already.

Follow a FaceStealer invitation through the cycle

Choose a view, then step through it. The first view follows the cycle the FBI describes, split into seven steps. The second lists the parts of the operation that appear outside a target's device. Each step separates what the sources report from a hypothetical outside response and the limit of that response.

How one victim leads to the next

Foothold

An earlier victim's device is already compromised

Reported evidence

The FBI says each cycle starts from a person whose device is infected. The operators scan it for wallets and seed phrases, read contacts and chat history, and secretly record the webcam while the person waits for a call to start (FBI FLASH).

Possible external intervention

None from outside. This stage happens on a private device.

Control boundary

Endpoint security and the person's own account security decide this stage. An external monitoring service has no view of it.

Hijack

The victim's messaging account is taken over

Reported evidence

The operators take over the victim's Telegram account or another encrypted messaging app and use it to reach the victim's contacts (FBI FLASH).

Possible external intervention

If the victim or a contact reports the takeover, a team could hypothetically log the account as a known source of lures and warn colleagues and partners who may receive messages from it.

Control boundary

The account is genuine, so there is no fake profile to detect. Recovery runs through the messaging platform, not a takedown request.

Invite

A tailored meeting request goes to a contact

Reported evidence

Using the stolen chat history, the operators propose a Zoom or Teams call to catch up, collaborate or discuss an investment. The FBI says they also use new and hijacked accounts on Telegram, LinkedIn, Slack and Discord for first contact (FBI FLASH).

Possible external intervention

Where the approach comes from a newly created profile that uses a real executive's or employee's name and photo, impersonation monitoring could hypothetically find it and request removal (Outtake executive protection).

Control boundary

Messages sent from hijacked genuine accounts inside private chats are not visible to outside monitoring.

Fake link

A lookalike Zoom or Teams link arrives shortly before the call

Reported evidence

The scheduling invite often has no meeting link. The link comes later from the trusted account and points to a typo-squatted domain such as uu01webzoom[.]us or teams.livemeet[.]us that hosts a copy of the meeting interface (FBI FLASH).

Possible external intervention

A reported link could start an investigation into related domains through shared hosting, registration patterns and other connections, so a team could block sibling domains and file abuse reports (Outtake brand protection).

Control boundary

The domains imitate meeting providers, not the targeted company, so watching the company's own names would not find them. Registrars and hosts decide whether to act.

Fake call

Recorded video stands in for the trusted contact

Reported evidence

The operators play the inviter's recorded webcam video, often with other recordings so the call seems to have several participants. The FBI says deepfakes of well-known industry figures, made from public photos and videos, are also used, though stolen genuine video still dominates (FBI FLASH).

Possible external intervention

Where synthetic video of a known leader circulates in public, earlier detection could support warnings to staff and partners and takedown requests (Outtake executive protection).

Control boundary

Video played inside an attacker-controlled meeting page is not public. Arctic Wolf says its own deepfake findings relied on provenance evidence, not appearance.

The fix

A fake audio problem leads the target to install malware

Reported evidence

The fake contact says the target cannot be heard and walks them through an update prompt, a script or a pasted terminal command that installs a cookie harvester, keyloggers, screen capture and a backdoor (FBI FLASH).

Possible external intervention

None from outside at this moment. A hypothetical earlier warning about active lures could make a target more cautious.

Control boundary

This stage is decided on the endpoint and by the person on the call. The FBI's advice is to refuse any in-call update, script or command and to update the meeting app separately.

Repeat

The new victim becomes source material

Reported evidence

The target's webcam has been recorded since they clicked the link, so the cycle starts again from their device and contacts. The FBI says thefts sometimes came months to more than a year later, and that an infected device can expose an employer's network or a maintainer's repository account (FBI FLASH).

Possible external intervention

Mapping a reported incident to related domains and accounts could help a team warn the people most likely to be approached next (Outtake platform).

Control boundary

Outside work does not remove malware, recover funds or restore a hijacked account.

What shows up outside the target's device

Domains

Typo-squatted meeting domains

Reported evidence

The FBI says the operators registered many domains that resemble Zoom and Teams. Arctic Wolf reports 98 high-confidence domains in the current infrastructure, 73 not previously reported (Arctic Wolf). Our November 13 RDAP check showed uu01webzoom[.]us registered on January 23, 2026.

Possible external intervention

Meeting-themed patterns and a reported link could seed hypothetical searches for related registrations, with evidence prepared for registrar and host abuse reports.

Control boundary

Registration data alone does not prove intent. Many lookalike names are parked or unused, and providers apply their own policies.

New accounts

New profiles on professional and chat platforms

Reported evidence

The FBI lists Telegram, LinkedIn, Slack and Discord as places where the operators create new accounts or use hijacked ones (FBI FLASH).

Possible external intervention

Fake profiles that copy a named executive or employee are a standard impersonation case for detection and removal requests (Outtake executive protection).

Control boundary

Profiles that invent a persona without copying a monitored person may not match any protected identity. Hijacked real accounts do not look fake.

Scheduling

Calendar invites without a meeting link

Reported evidence

The FBI says Calendly or similar invites often arrive without a link, with the link sent from the trusted account shortly before the call to limit exposure of the infrastructure (FBI FLASH).

Possible external intervention

Internal guidance can treat a link-free invite followed by a late link as a warning sign. Any scheduling pages tied to reported activity could be added to an investigation.

Control boundary

This pattern sits in private email and chat. It keeps infrastructure out of public view until shortly before each call.

Deepfakes

Synthetic video of known industry figures

Reported evidence

Arctic Wolf says the operators screen-record YouTube interviews and conference talks to capture likeness and voice, and flagged 349 videos as likely modified as of November 6 (Arctic Wolf).

Possible external intervention

A leader whose public talks are being reused could hypothetically be alerted when synthetic clips or impersonating accounts surface publicly, and partners could be warned.

Control boundary

Public talks cannot be taken down because someone copies them, and footage used only inside fake calls may never surface.

New lures

Fake document viewers and update pages

Reported evidence

Arctic Wolf describes new pretexts including shared financial documents, a fake Chrome PDF Viewer update, a fabricated media player called NovaStream, identity checks, hardware wallet access and security audits (Arctic Wolf).

Possible external intervention

Public lure pages that copy a product or brand could be found by visual comparison and reported, where the imitated owner or a targeted organization is monitoring for them (Outtake brand protection).

Control boundary

Lures change quickly, and pages served only to invited targets may not be reachable by outside scanning.

Seizure

What the November 12 disruption changes

Reported evidence

The DOJ says cloud-based infrastructure was seized and more than 900 victim videos recovered under warrant. It has not said how much infrastructure remains active (DOJ).

Possible external intervention

Teams could use published indicators and the FBI's IC3 reporting channel, keyword FaceStealer, to check past contacts and report new approaches (FBI).

Control boundary

A seizure removes some assets, not the operators. Arctic Wolf's reporting suggests new infrastructure and lures continue to appear.

What outside monitoring cannot reach

Most of FaceStealer's decisive moments are invisible from outside. The takeover of a messaging account, the private invitation, the fake audio error, the pasted command and the malware all happen in private channels or on the target's device. Endpoint security, account security on messaging apps and wallet controls are where those steps are stopped. The FBI's own recommendations focus there: join calls through the official app with an invite code, wait to turn on the webcam until you can see and talk with the inviter, refuse in-call update prompts, confirm invitations through a second channel, protect messaging accounts with multifactor authentication and login alerts, and keep larger crypto balances off connected devices. People who suspect a compromise are told to warn contacts and employers and to report to IC3 with the keyword FaceStealer (FBI FLASH).

An external service would also be limited by the operators' habits. The FBI says scheduling emails often omit the meeting link, which keeps infrastructure out of view until shortly before each call. Arctic Wolf reports frequent changes to the visible lures. A takedown removes a page or an account. It does not recover stolen funds, restore a hijacked account or clean an infected machine. For most organizations, the imitated meeting brands belong to someone else, so registrars and hosts will judge reports on the abuse evidence rather than on the reporter's trademark rights.

Readers should treat the dollar figure and the video and target counts as government and vendor statements, and expect more detail if charges or forfeiture filings follow. Security teams that want background on the related evergreen risks can read Outtake's guides to executive impersonation and deepfake and synthetic media detection.

Frequently asked questions

Is FaceStealer the same operation as the Binance theft?

No. The Justice Department announced both actions on November 12, but it attributes the Binance theft of on or about October 29 to a group it calls TraderTraitor and treats FaceStealer as a separate scheme. Reuters also described the two clusters as distinct (DOJ, Reuters via Investing.com).

Did the FBI say more than 900 people lost money?

The FBI's text says it identified and seized more than 900 victim webcam videos, and the advisory's title refers to a webcam library of more than 900 tech workers. Some coverage, including the BBC, described this as more than 900 victims. The FBI has not said how many of the people recorded lost funds. Arctic Wolf separately counts 677 people for whom material was prepared, which it says is not a count of confirmed compromises (FBI FLASH, BBC, Arctic Wolf).

How can someone spot a FaceStealer meeting invitation?

The FBI advises joining calls through the official Zoom or Teams app with an invite code instead of a web link, checking calendar invites for missing or late meeting links, confirming any meeting through a different channel, waiting to turn on the webcam until you can see, hear and talk with the person who invited you, and refusing any update, script or command offered during a call (FBI FLASH).

Why are open-source maintainers a target?

A maintainer's access can reach everyone who installs their code. The FBI says a maintainer of a JavaScript library with tens of millions of weekly downloads was infected through a fake Teams call earlier this year and that the operators used his access to create malicious dependencies. The FBI calls it a failed attack stopped by quick detection and response, and it reports attempts against maintainers on npm and of Rust crates (FBI FLASH).

What could an external digital risk protection service do about this kind of scheme?

It could help with the public parts: lookalike meeting domains, fake profiles that copy real executives or staff, and synthetic video that surfaces in public, along with evidence for abuse reports. It cannot see private chats, stop malware on a device, restore hijacked accounts or recover stolen funds (Outtake executive protection, Outtake brand protection).

Sources