
The FBI, the U.S. Secret Service and IRS Criminal Investigation have attributed the Binance theft, reported at more than $6 billion, to North Korea's TraderTraitor. The attribution names no exchange bug. The group's published playbook still gives defenders a test plan: TraderTraitor job lures aimed at developers, followed by a poisoned infrastructure toolchain.
Key takeaways
- The FBI, Secret Service and IRS-CI attributed the Binance theft to TraderTraitor (also tracked as Jade Sleet and UNC4899) but published no root cause or entry vector.
- Reported totals range from $6.3 billion to $6.43 billion. Either way, this is the largest crypto theft on record, ahead of Bybit's $1.46 billion in February 2025.
- $2.85 billion is reportedly frozen, but frozen is not recovered, and native BTC and ETH have no issuer that can lock a balance.
- TraderTraitor's documented playbook is fake recruiters, AI-polished personas and trojanized Terraform providers delivering the FLATROOF stealer and ROOFDECK backdoor.
- Crypto firms should load the FBI wallet list into screening on every supported chain now, then test whether a job lure can reach cloud or CI credentials.
What did the FBI attribution actually say?
It says who took the money, not how they got in. Coverage dated 1 November 2026 reports that the FBI, working with the U.S. Secret Service and IRS Criminal Investigation (IRS-CI), attributed the theft of virtual assets from Binance to TraderTraitor. TraderTraitor is a North Korea-aligned cluster that has gone after Web3 and cryptocurrency firms since at least 2020.
Vendors track the same cluster as Jade Sleet (Microsoft’s label), Pressure Chollima, PUKCHONG, Slow Pisces and UNC4899. If several of those names show up in your threat-intel feed, they describe one adversary. Deduplicate detections and reporting accordingly.
The FBI also described where the money is going. The actors are converting stolen assets primarily into Ethereum. The bureau expects the next stage to run through decentralized finance protocols and instant swap services:
“It is expected these assets will then be further laundered by converting them using DeFi and instant swap services as they prepare them for further laundering stages.”
The bureau published wallet addresses it says are being used for laundering. We don’t reproduce them here. Pull them from the FBI release so you work from the authoritative, current version.
Four things are missing from the statement:
- a root cause
- an initial-access vector for the exchange
- any confirmation that the tooling described below was used against Binance
- any technical timeline
If you read our earlier coverage of the Binance incident, keep that distinction in mind. This attribution does not confirm any specific entry path. Anyone who briefs your board otherwise is speculating.
How big is the theft, and how much is actually frozen?
Nobody agrees on the exact number, and it tells you nothing about your own exposure anyway. The Hacker News ran $6.43 billion in its headline and “more than $6 billion” in its body. Infosecurity Magazine and BleepingComputer used $6.3 billion. Token prices move by the hour, so a USD total for a multi-chain theft drifts with the timestamp. Treat the precise figure as unconfirmed.
The other numbers are consistent across reports:
- Spread: the funds were taken across 11 blockchains, with 82% on Ethereum and Bitcoin.
- Scale: any of the reported totals makes this the largest crypto theft on record, ahead of the $1.46 billion taken from Bybit in February 2025.
- History: DMM Bitcoin, Bybit, KelpDAO and Bitget are named as earlier heists attributed to the group.
Elliptic estimates that North Korea-linked actors have stolen over $6.5 billion in crypto assets in recent years. The firm also says “this single incident almost doubles the total cryptoassets stolen by the regime.” Those two lines don’t reconcile cleanly. Cite them as Elliptic’s characterization rather than doing the math yourself.
A freeze is not a recovery
Reporting puts $2.85 billion as frozen on-chain, with Tether, Circle, Ripple, Arbitrum and Optimism named. Those are different levers. Tether, Circle and Ripple issue tokens and can lock their own assets. Arbitrum and Optimism are Ethereum layer-2 networks, which can act only where their operators retain that control.
Frozen is not seized, recovered or returned to Binance. Native BTC and native ETH have no issuer that can lock a balance, and the FBI says the actors are converting into Ether. Every hour of laundering shrinks the share that can still be frozen.
Binance’s first day back
Binance paused withdrawals after the incident and kept deposits open, describing them as unaffected. Phase one of withdrawals reopened between 06:00 and 09:00 UTC on 1 November. The exchange later restored all 15 networks while it kept testing “remaining networks and assets.”
Between 06:00 and 15:30 UTC it processed $4.649 billion in withdrawals and $1.01 billion in deposits, a net outflow of about $3.64 billion. That shows Binance absorbed a liquidity test. It says nothing about what an attacker could take from you. Scope your testing around your own blast radius, not someone else’s headline.
How do TraderTraitor job lures reach a developer?
They reach a person, not an exchange API. The opening move is a recruiter. Microsoft says Jade Sleet used “tailored domains, social media profiles, and industry-specific knowledge to pose as recruiters for a spoofed AI-powered blockchain intelligence company.” The actor offered software engineering roles and scheduled interviews with anyone who replied.
The company was fake, but it was constructed in depth: dedicated domains, layered personas, and enough industry knowledge to survive a technical conversation. Microsoft says the actor ran similar campaigns through two more front companies in a short window, with comparable infrastructure and the same command of the industry.
The target list is the detail most write-ups skip. The lures went to people in cybersecurity, AI, blockchain and talent recruitment. Recruiters open documents and links from strangers for a living. Security and AI engineers hold privileged access and tend to trust their own judgment about “obviously safe” tooling.
The method mirrors two other North Korean campaigns, the Lazarus Group’s Operation Dream Job and Famous Chollima’s Contagious Interview. All three use a fake employer, a real conversation and a technical task that executes code. Treat them as one class of threat your hiring and engineering workflows must resist, whichever cluster sent the email.
The reporting says AI helps the group craft lures, build fake personas and evade detection. That strips out the tells recruiter training relied on: awkward grammar, thin LinkedIn histories, generic job descriptions. A model can write fluent outreach and a role that matches your engineer’s public GitHub. None of this suggests an autonomous agent operated inside Binance.
The chain has worked against an exchange before. The FBI and Department of Justice attributed the 2024 DMM Bitcoin theft to TraderTraitor. Their account ran from a LinkedIn recruiter lure, through a malicious Python pre-employment test, to the compromise of an employee at Ginco, a wallet-software provider. That precedent does not tell us how Binance was breached.
What does a trojanized Terraform provider actually do?
It runs attacker code with whatever privileges your Terraform run already has. SentinelOne, Zscaler ThreatLabz and Microsoft have all reported TraderTraitor distributing malware through poisoned providers. The reporting names no specific provider, registry entry or version, so take IOCs from those vendors’ own reports.
The reported chain has three stages. We give names and impact only:
- Bash loader. The poisoned provider delivers a loader that identifies the victim’s operating system and fetches a matching payload for macOS, Linux or Windows.
- FLATROOF. A cross-platform implant that uses a Python-based information stealer to siphon sensitive data from the host.
- ROOFDECK. A backdoor dropped at the end of the chain that gives the operator complete remote control.
This works because teams treat providers like configuration. They are executable plugins that Terraform downloads and runs. HashiCorp’s dependency lock file records provider checksums, which catches a package that changes later. It cannot catch a package that was malicious on first download: the lock file faithfully pins the malware.
Look at what usually sits on a machine that runs terraform plan:
- cloud credentials for the target accounts
- read access to state files, which often hold secrets in plain text
- CI tokens and SSH keys
- a browser with live sessions to the cloud console
An infostealer on that host doesn’t need an exchange zero-day. It needs one engineer who accepted a take-home exercise.
What we find on the developer laptop
In our assumed-breach engagements, the developer laptop is rarely where the story ends. We most often find long-lived cloud keys in a local profile, a state bucket the engineer can read, and a CI token with write access to a deploy pipeline. Together they turn one workstation into production-adjacent access.
Detection checks
- Does your EDR raise an alert when a terraform process spawns bash, curl, wget, Python or PowerShell?
- Can you correlate provider downloads with DNS, proxy and CI artifacts?
- If a developer machine is suspected, do you rotate from a known-clean system? Cover cloud tokens, CI secrets, SSH keys, signing credentials, API keys and browser session tokens. Resetting only the obvious password leaves the persistence paths that matter.
What should crypto firms do with the FBI wallet list?
Load the addresses into screening today, then confirm the blocks actually fire. This matters most for exchanges, e-money institutions and crypto-asset service providers (CASPs), including the many authorized by CySEC in Cyprus under MiCA. Instant swap services and DeFi on-ramps are exactly where the FBI expects the funds to surface.
- Cover every chain you support. Load the list into deposit and withdrawal screening and analytics rules for all networks, not just Ethereum. Timestamp each version so you can show when blocking started.
- Screen exposure, not only direct matches. The funds will pass through swaps and intermediate wallets. A deposit one or two hops from a listed address should trigger a hold and review, not an automatic credit.
- Don’t treat the list as complete. OFAC states that its published address identifiers are not exhaustive. Sanctions exposure can arise through addresses nobody has listed yet.
- Pre-write the hold-and-report path. Decide in advance who freezes a flagged deposit, who files with the financial intelligence unit (MOKAS in Cyprus) and who talks to law enforcement.
- Test the rule, not the spreadsheet. Seed synthetic entries into a staging copy of your screening rules and confirm that deposit, withdrawal and internal-transfer paths all stop. Address-format quirks and checksum casing break more blocklists than missing data does.
- Track statuses separately. Frozen, seized, recovered and returned are four different states. Keep transaction hashes and exact asset denominations, and report each state on its own to management and regulators.
How do you scope a recruiter-to-laptop-to-cloud test?
Scope the published tradecraft as three separate controls, and assume the exchange vector stays unknown. Replaying the Binance theft is not the goal. The goal is to measure how far the same opening move gets inside your company.
People. Put staff who take external interviews or talk to recruiters in scope: security, AI, blockchain and recruiting teams, the same functions Microsoft saw targeted. A social engineering assessment can run a job-lure pretext under authorization and record who engages, who runs the exercise and on which device.
Endpoints and toolchain. Test personal and corporate devices separately, then test the infrastructure-as-code path on its own:
- where providers and modules are sourced from
- whether an internal mirror or allowlist exists
- whether lock file changes get reviewed
- whether EDR sees unusual child processes from Terraform
Blast radius. An assumed breach assessment starting from a developer foothold answers the board’s question. From one laptop, what can an attacker reach in cloud accounts, CI pipelines, secret stores and signing infrastructure? How long does it take, and does anyone notice?
For EU financial entities this is not optional homework. CASPs authorized under MiCA fall under DORA, which requires a resilience testing program and, for entities designated as significant, threat-led penetration testing. A TraderTraitor scenario is a defensible threat intelligence input for DORA TLPT: documented, attributed and explicitly aimed at this sector.
A passing result has five parts:
- The fake recruiter fails out-of-band verification.
- Exercise code runs only in a disposable sandbox with no corporate credentials.
- Untrusted code cannot run at all on personal devices that hold live work sessions.
- A rogue provider cannot be fetched from outside your allowlist, and the laptop holds no long-lived production credentials.
- Your SOC raises an alert before the assumed-breach team reaches a signing key.
Anything short of that is a finding with an owner and a date.
What is still unknown?
Most of what matters for root-cause analysis is still unpublished:
- How Binance was breached. There is no published vector.
- The tooling link. Nobody has confirmed that the Terraform, FLATROOF or ROOFDECK chain played any part in this incident.
- The exact figure. The USD total is unresolved, with $6.3 billion and $6.43 billion both in circulation.
- The frozen funds. Nobody knows yet how much of the $2.85 billion will move from frozen to returned.
- Binance’s own report. The exchange has not said whether it will publish a technical post-incident review.
Until those answers arrive, don’t anchor your security program on a guessed exchange bug. Anchor it on the parts of the chain that were actually published: the recruiter, the developer tool and the credentials that sit behind them.
FAQ
Who did the FBI blame for the Binance crypto theft?
The FBI, with the U.S. Secret Service and IRS Criminal Investigation, attributed it to TraderTraitor, a North Korea-aligned group that has targeted cryptocurrency and Web3 firms since at least 2020.
What is TraderTraitor also known as?
Security vendors also track the group as Jade Sleet (Microsoft), Pressure Chollima, PUKCHONG, Slow Pisces and UNC4899.
Did the FBI say how Binance was hacked?
No. The attribution names the group and describes laundering into Ethereum through DeFi and instant swap services. It gives no root cause or initial-access vector for the exchange.
Is the Binance hack bigger than the Bybit hack?
Yes. Reported totals of $6.3 billion to $6.43 billion are several times the $1.46 billion stolen from Bybit in February 2025, which makes Binance the largest crypto theft on record.
How much of the Binance stolen crypto was frozen?
About $2.85 billion has reportedly been frozen on-chain, with Tether, Circle, Ripple, Arbitrum and Optimism named. A freeze is not a seizure or a return of funds to Binance.
How do North Korean fake job offers install malware?
A fake recruiter builds rapport, then hands the target a technical task that runs code. TraderTraitor has used trojanized Terraform providers that deliver a Bash loader, the FLATROOF stealer and the ROOFDECK backdoor.
If you want to know how far a fake recruiter gets from one of your engineers’ inboxes, agent::offense runs authorized job-lure scenarios and developer-foothold assumed breach assessments scoped for MiCA and DORA obligations.