Skip to content
THREATWIRE

2026-09-04 · record inc_0006 · updated 2026-10-19

Pragma and Vesu: a wrong USDT conversion triggered 47 liquidations

A stale conversion route and a two-source median made major prices look halved for minutes on Starknet; Vesu liquidated 47 positions.

PragmaVesuStarknetComponent: Pragma publisher software, USDT/USD aggregate; Vesu liquidations

Event
Official statement
Root cause
Official statement
Loss
Official statement

Current status

Assets recovered; curator refunds pending.

Timeline

  1. 2026-09-04 04:07:52 UTCUSDT/USD aggregate becomes incorrect.
  2. 2026-09-04 04:08:16 UTC47 liquidations begin; window ends about 04:10.
  3. 2026-09-04 07:59 UTCSDK 2.13.1 hotfix.
  4. 2026-09-13Vesu announces completion of asset recovery.
  5. 2026-09-22Oracle administration moved to a 3-of-5 multisig.

Root cause

Per Pragma, a registry mapped USDC to bridged USDC.e, producing a bad USDT price; freshness filtering left two sources; the publisher then reused that price to convert other feeds.

Basis: official post-mortem · verified by us against that source

Classes: Oracle and spot-price manipulation, Liquidation and collateral logic

Impact: loss and recovery

47 liquidations across 42 borrower wallets in seven pools.

Reported loss

$1.4M

USD basis: Pragma reconstruction: borrower equity lost plus pool bad debt. Collateral seized was $3.08M, a different measure.

Recovered or returned

$1.33M (partial)

Vesu reports 95% recovery at 11 September prices; curator refunds in progress.

Loss and recovery are separate figures. They are never netted, and amounts in different currencies are never added together.

Addresses and transactions

No addresses or transaction hashes have been verified for this record. We do not publish unverified hashes.

Public reproduction file

No public reproduction file is recorded for this incident.

Remediation and defensive indicators

  • SDK hotfix, measured USDT conversion, four-source default for USDT and USDC, 3-of-5 multisig. (reported)

Defensive lessons from the class

  • Estimate what it costs to move each price and compare it with what that price controls.
  • Require a minimum number of genuinely independent sources and reject answers below it.
  • Limit exposure to thin collateral.
  • Add circuit breakers on abnormal moves.

Sources

Similar incidents

Read the full written analysis

Method, uncertainty and corrections

Still unknown

  • Final user loss after refunds
  • Rollout verification across publishers

Where sources disagree

  • Recovery rate uses different prices from Pragma's loss estimates.

Written by ThreatWire editorial. Facts attributed to the sources listed; figures are estimates and not independently verified unless marked on-chain. Published 2026-10-19, last updated 2026-10-19. Figures are estimates from public post-mortems and named analysts; not independently verified unless marked as confirmed on-chain.

No corrections recorded since publication.