US says North Korean hackers stole $400m via fake video calls

The US Department of Justice said on 12 November 2026 that North Korean hackers stole more than $400 million in cryptocurrency over 18 months using a scheme the FBI calls FaceStealer. The attackers used accounts taken from earlier victims to invite new targets to video meetings, then played recorded webcam clips of people those targets knew, and used AI deepfakes of well-known figures. The FBI seized more than 900 webcam videos under a federal warrant, and the DOJ said it took down the scheme's cloud infrastructure. North Korea denies involvement. For businesses that automate payments, approvals and system access, the lesson is that a familiar face on a video call is not proof of identity. Any request to pay, change bank details or grant access should be confirmed through a contact route the company already holds.
In short
- North Korean hackers reportedly stole over $400 million in crypto over 18 months, according to the US Justice Department.
- The attack uses hijacked accounts, replayed webcam clips and AI deepfakes to get victims to install malware.
- A familiar face or voice on a video call is not proof of identity for payments or access.
- Set a callback on a number already on file and a second approver for any money or bank change.
- Automation can enforce these checks, but it cannot judge whether a video is real.
What happened
According to the US Department of Justice, North Korean hackers stole more than $400 million in cryptocurrency over the past 18 months using a social engineering scheme the FBI calls FaceStealer. The department announced the disruption on Thursday, 12 November 2026. Its press release says the scheme targeted people working in the cryptocurrency and Web3 sectors, and that the hackers have increasingly gone after people who build and maintain open-source software.
The method works in steps. Attackers take over an account that belonged to an earlier victim and use it to invite a new target to a video meeting. Once the target joins, the attackers record the target's webcam and play a prerecorded video of someone the target knows, so the meeting appears to include trusted colleagues. For well-known figures, they use AI-generated deepfake video. According to The Block's account of the DOJ announcement, the callers then fake a technical problem and get the victim to install malware that can steal crypto or reach other systems.
The Justice Department said the FBI seized more than 900 webcam videos of victims under a federal warrant, and that it seized cloud-based infrastructure used in the scheme. The same day, the department obtained court authorisation to seize or restrain more than $1.2 billion linked to a separate theft from the Binance exchange, according to the department's release. The DOJ attributes the activity to North Korean groups tracked as Sapphire Sleet, BlueNoroff and APT38. North Korea has denied involvement. Decrypt reports that Pyongyang called the FBI's finding a political provocation.
Why it matters for businesses that automate work
The scheme does not break into a video platform. It uses trust in a face, a voice and a meeting invitation. The FBI's description is blunt. Assistant Director in Charge Patrick Grandy of the FBI's Los Angeles Field Office said FaceStealer is capable of deceiving even the most experienced technical and security-conscious computer user.
This matters to automation because many business workflows already act on requests that arrive by message, call or meeting. A payment run, a vendor bank change, a password reset or a new user account can each be triggered by someone who looks like a colleague. If an automated process accepts that request as authority, the attacker only needs to convince one person to join the call or click the link.
What changes in practice for a business that automates work
The FBI material cited in the DOJ announcement focuses on the attacker's method, not on whether a person can spot a fake. The FBI's own description of the deepfakes makes the point. Quality only needs to be good enough to pass as webcam video. That is why a company should not depend on spotting the fake during a call. It should instead decide in advance which requests need a separate check. The table below sets out controls that do not depend on what the video looks like.
| Control | What it does | Who owns it |
|---|---|---|
| Callback on a number already on file | Ends any payment or bank detail change requested on a call, chat or video, and confirms it on a number held in company records. | Finance or accounts payable lead |
| Second approver for payments | Requires a second named person to approve any transfer above a threshold the owner sets. | Owner or finance manager |
| Rules for meeting links | Treats any meeting invitation about money, access or confidential files as unverified until confirmed by callback. | Operations head and IT |
| Review of automated accounts | Checks which accounts and tools can start payments, add users or export files, and removes unused access. | IT or the software owner |
| Evidence retention | Keeps call logs, meeting links and messages when a suspicious request is received, for the FBI's complaint process at ic3.gov. | Operations head |
Automation can enforce these steps. It cannot judge whether a face on a screen is real. An automated workflow can hold a payment until a person confirms it through the callback route, send the request to a named approver, and record what was checked. It cannot decide that a video call is genuine, and it should not be set up to act on a video or voice instruction alone. The trade-off is speed. Urgent payments will wait for a callback, and a business should decide in advance which cases are urgent enough to skip a step. In most cases, the right answer is that they should not skip it.
What to watch next
Three things are worth tracking. First, the 900-plus seized recordings may lead to further court filings, but the coverage of the announcement does not mention individual charges. Second, North Korea's denial stands, so attribution rests on US agencies' findings for now. Third, the DOJ's warning about open-source developers suggests that attackers may move from payments to software supply chains. Businesses that depend on code repositories should review who can publish updates and how those accounts are protected.
How AiStaffo would automate this
AiStaffo would connect an AI worker to the places where payment, vendor bank change and access requests arrive, such as the accounts payable inbox, the finance tool and the user access list. The worker would hold any request to pay, change bank details or add a user, then open a callback task that names the contact to ring from the company's own records. It would not act on the meeting or the video itself. The business owner or a named approver would still confirm each held request and decide whether to release it. The first step is a free review of which requests in your own workflows can move money or grant access. Book a free automation audit.
Questions people ask
What is the FaceStealer scheme?
How much did the FaceStealer scheme steal?
Did North Korea respond to the FBI claims?
How can a business avoid being tricked by a deepfake video call?
Can automation stop deepfake video call fraud on its own?
What should a business do if it thinks it has been targeted?
Book a free automation audit
Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.

































