# Blockstream Traces the $320 Million Liquid Theft to a Proof-Verification Cache in Its Own Node Software

Nearly 4,000 unbacked Liquid bitcoin were minted and 3,996 bitcoin were withdrawn in a single block. About 598 bitcoin remain with an attacker who calls himself a security researcher.

- Published: 2026-09-24T05:46:16.721Z
- Canonical: https://polylog.news/crypto/2026-09-24/blockstream-traces-the-320-million-liquid-theft-to-a-proof-v
- Publisher: Polylog (Crypto desk)
- Section: crypto
- Sources: [Blockstream Blog](https://blog.blockstream.com/liquid-network-security-incident-assessment/), [Rekt News](https://www.rekt.news/)

Blockstream has published its [security incident assessment](https://blog.blockstream.com/liquid-network-security-incident-assessment/) of the September 6 attack on Liquid, the bitcoin sidechain it operates for exchanges and trading firms. The report, along with independent analyses collected by [Rekt News](https://www.rekt.news/), points to the same failure. Liquid nodes cache the result of verifying range proofs, the cryptographic checks that confirm confidential transaction amounts are valid without revealing them, and that cached result was reused in a context it did not actually cover.

The consequence was an unbacked mint. The attacker created close to 4,000 Liquid bitcoin without depositing any collateral and withdrew 3,996 bitcoin through the peg, cutting operating reserves from 4,205 bitcoin to 197. Blockstream patched the bridge nodes within hours, block production stopped for roughly four days, and the network resumed on September 10. Roughly 3,400 bitcoin came back almost immediately. The remaining 598.5 bitcoin have not, and the person holding them has asked for a ten percent payment in exchange for returning them. Blockstream has said it will not pay to recover stolen funds.

The defect was not in application code. It sat in an optimization inside Elements, the open-source node software Liquid inherits from Bitcoin Core, so audits aimed at contract logic never covered the place the funds actually left the system. The same pattern has surfaced this year in oracle forwarders and validator signature sets: the component did exactly what it was written to do, and the specification behind it was wrong.

Liquid also demonstrated the trade-off built into its design. A federation of functionaries holds the peg keys and can coordinate a halt, which is how 3,400 bitcoin were stopped and how a patch reached every bridge node in under ten hours. A permissionless chain has no such option. Users who chose Liquid for fast settlement accepted that a small, identifiable group controls both the failure and the remedy.

## What this means

Exchanges and market makers holding Liquid bitcoin carry the residual loss through the peg, because reserves briefly fell to 197 bitcoin against nearly 4,000 in circulating claims, and the backing was restored only through a voluntary return and a coordinated halt. The wider exposure reaches every chain running Elements-derived code, and it tests Blockstream's institutional sales case, which rests on the argument that a known federation is safer than an anonymous validator set.

## What to watch

- Whether the remaining 598.5 bitcoin move on-chain or are returned, which will show whether refusing to pay a self-described white hat recovers funds or simply keeps them frozen in place.
- Whether other chains built on Elements or on similar confidential-transaction code publish patches for the same caching path, which would show the defect is shared rather than specific to Liquid.
- Whether exchanges keep issuing and holding Liquid bitcoin at previous volumes, the clearest measure of whether institutional users accept the federation model after seeing it halt a chain.