Liquid Network
Why it happened
Attack class: Logic errorLiquid nodes cached proof checks with a cache key that could be matched by crafted bytes, so an invalid proof was accepted and unbacked coins were later redeemed for real bitcoin.
What would have stopped it
- Peg-out size and velocity limits
- Reserve-versus-supply monitoring that pauses automatically
Not scored before the exploit
BlackHart had not scored Liquid Network before this exploit.
Someone exploited a flaw in Elements, the software behind the Liquid sidechain, to create about 4,000 L-BTC that had no bitcoin behind it. They sent it to SideSwap's peg-out service, which converted it to real bitcoin, and the Liquid Federation released about 3,996 BTC, roughly $320 million at the time. The actors posted that they were white hats, and about 3,400 BTC was returned the next day, leaving about 600 BTC outstanding per Blockstream's last report. Blockstream says no keys were compromised and other Liquid assets were not inflated, though the network was paused while it recovered.
On Liquid, the attacker submitted a transaction that Elements nodes accepted as valid, creating about 4,000 unbacked L-BTC. The attacker then used SideSwap's peg-out service, and the Federation released about 3,996 BTC on Bitcoin. Later, about 3,400 BTC was returned to the Federation peg wallet.
Attack flow
All 4 on-chain stepsThe attack in 3 steps. Step 1, Unbacked mint: Liquid sidechain (source). Step 2, Peg-out: SideSwap peg-out service (hub), about $320M reached this step. Step 3, Reserve drain: Actor wallet (exit).
Step 1: Unbacked mint
- SourceLiquid sidechainabout 4,000 L-BTC createdno bitcoin behind it
Step 2: Peg-out
- HubSideSwap peg-out servicereceived about 4,000 L-BTCburned it and requested BTC
Step 3: Reserve drain
- ExitActor walletabout 3,996 BTC receivedabout 3,400 BTC returned
Why each step worked
- Range proof cache accepted forged matchcache key built from fields joined without length markers
- Earlier cache bug fixed but not yet visiblepublic fix revealed the code of the second flaw on September 1
- Online peg-out signing keymissing controlpayouts forwarded automatically
- No peg-out size or velocity limitmissing controlone request drained most reserves
- No supply-versus-reserve alarmmissing controlhalt came hours after the peg-out
Full forensic detail
Step-by-step reconstruction, root cause, counterfactuals, remediation, and disclosure timeline.
Exploit anatomy
Root cause
Liquid's nodes cache the result of range proof verification so they do not repeat expensive checks. Per Blockstream's incident report, an earlier version of the cache key left out parts of the data a proof is tied to, and the fix for that issue joined fields together without length markers. That let the attacker build a proof and script whose bytes lined up with a previously cached valid entry, so the node returned 'valid' without actually verifying it. The result was L-BTC that nodes accepted as real but that had no bitcoin reserve behind it. The peg-out then worked as designed: SideSwap's peg-out service burned what looked like normal L-BTC and the Federation released real bitcoin. Blockstream states that no Federation or SideSwap key was compromised. It also says SideSwap's peg-out authorization setup kept its signing key online, paid out automatically and applied no size or velocity limit, which meant a single large peg-out passed through unchecked.
Prevention analysis
A cap on how much bitcoin one peg-out request could release would have turned a reserve-draining event into a small one.
An automatic halt when L-BTC supply exceeded reserves would have stopped the peg-out within minutes, not hours.
Keeping the peg-out authorization key offline, as the Federation charter requires, would have forced a review before a record-size release.
Similar incidents
A voucher on a bitcoin-backed chain was created without matching reserves, and the bridge honored it for real bitcoin.
Forged proofs were accepted as valid by on-chain verification, releasing funds that had no real backing.
Remediation
Timeline
Get your protocol scored across 12 dimensions, or request ongoing coverage.