BLOCKCHAIN LAB BRIEFING · BITCOIN
A World Bank Working Paper Estimates a Bitcoin Leak in Aid. The Range Is Not a Seizure.
NBER working paper 35655 estimates an implied leak into crypto around World Bank disbursements. The cents-per-dollar figure moves with the hop count the authors assume.
1 October 2026
01
What happened
On 1 October 2026 a widely shared post said researchers had found a spike in bitcoin activity when foreign aid goes out, and that somewhere between 2 and 6 cents of every World Bank aid dollar was siphoned into crypto wallets. The post did not link the paper in the text that was retrieved. The chart, for most readers, was the argument.
The underlying document is NBER working paper 35655. The authors estimate an implied leak by watching exchange activity around disbursements. They are explicit that exchange ledgers do not reveal a turnover multiplier, so they map hops to a net figure. At the two-to-five hops that forensic practice often treats as typical, they arrive at roughly 2 to 6 cents per aid dollar. On the tranche arrivals used for that ratio, they put a dollar total of about $1.7 billion to $4.4 billion.
The word “siphoned” is the poster’s. The paper’s claim is an estimate under a hop assumption, not a set of wires the authors watched leave a project account.
02
Why it matters
Aid leakage is a governance problem with a measurement problem sitting on top of it. If a share of disbursements can be seen moving toward crypto venues, ministries and banks that touch those programmes have a monitoring question: which wallets, which hops, and which assumption turns a flow into a cents-per-dollar figure.
For builders of compliance and payment systems, the paper is a reminder that a public ledger can show a path and still not show intent, a beneficiary, or a loss that a court would recognise. The estimate is not a seizure total.
03
The operating layer
Change the hop count and the cents change. That is the sentence a short post leaves out. A reader who treats 2 to 6 cents as a measured theft, or who inflates it to 2 to 6 percent, is no longer reading the paper.
Some replies did exactly that, turning cents into percent. Others asked for the link. The correction is in the method, not in the argument about whether theft of aid is serious.
04
What is verified
The working paper exists, and the range quoted in the post matches the authors’ implied figure at a stated hop band. The dollar total above is the paper’s, on the tranches it used.
Blockchain Lab has not re-run the clustering, and has not identified any recipient government. This briefing is not a comment on a particular country.
05
What remains unclear
The post does not say which exchange clusters, which corridor, or which years a non-specialist should treat as the result. Without the paper, the chart cannot be checked.
A leak estimate is not a finding that bitcoin caused the diversion. The same funds could have moved through cash, trade misinvoicing or a correspondent bank and never touched a chain.
06
The catch
Public attention landed on a round number. The research landed on a range that moves when the forensic assumption moves. Quoting the top of the range as a fact, or treating the figure as money already recovered, misstates the work.
Nothing here is a reason to hold or sell bitcoin. Nothing here is a loss figure for a named project.
WATCH
What builders should watch
- 01Whether the authors or the World Bank publish a response that fixes the hop assumption.
- 02Whether a later version separates detected flows from attributed theft.
- 03How development banks describe wallet monitoring in the programmes that actually disburse.
BOTTOM LINE
The paper is a serious estimate of a possible leak, with the uncertainty written into the method. The short version that travelled is stronger than the estimate. Read the hops before you repeat the cents.
Sources
Blockchain Lab uses public social posts as reporting leads, not as proof. Every published briefing is assessed against primary sources, available documentation and relevant technical context. Social engagement is not used as evidence of the underlying claim.