Skip to content

exploit

Premia vaults on Arbitrum drained ~$437k via empty option-pool and flash-loan routing

Attacker drained roughly $437k across two txs from six Premia option vaults on Arbitrum, using empty option pools, flash loans and option purchases to park collateral then withdraw it as liquidity.

by 5 min read

Premia Labs' options-vaults protocol on Arbitrum was drained of roughly $437,000 in collateral on October 31, 2026, after an attacker chained empty option-pool creation, flash loans and option purchases to route vault collateral into pools from which it was later withdrawn as liquidity. The on-chain alert was Blockaid's exploit-detection post, which first flagged the activity as "ongoing" and estimated ~$400,500 across two draining transactions. The Defiant later put the total at about $437,000 across six vaults; ChainCatcher relayed Blockaid's technical account.

What happened

Per Blockaid, the attack moved vault collateral into option pools before withdrawing it as liquidity. The sequence, reconstructed from the Blockaid and secondary accounts:

  1. A setup transaction deployed six option-pool contracts.
  2. The attacker then opened empty pools, took flash loans against them, and bought options priced in a way that locked the vaults' collateral into the pools the attacker had just created.
  3. The attacker then withdrew that collateral as liquidity from the new pools.
  4. Two draining transactions settled the funds out, leaving the six underlying Premia option vaults with "little direct collateral remaining," per Phemex News's reconstruction.

Chain: Arbitrum, confirmed across sources. Blockaid said the attack was still ongoing when it published; the ~$437k figure in The Defiant is the later, higher post-hoc total.

As of writing, no attacker address or specific drain-tx hashes have been published by Blockaid, The Defiant, Premia or ChainCatcher's secondary account — only the two-tx structure and the six-pool fingerprint. Premia Labs has not published a post-mortem.

Mechanism — the shape of the primitive

Premia runs underwriter vaults that take user deposits (WBTC, WETH and ARB are the headline collateral assets on Arbitrum) and quote prices on calls and puts across live option pools. The quote path depends on pool liquidity and spot-vs-strike geometry.

The attack appears to abuse that quote path. By spinning up six brand-new pools with the right strikes and by flash-loaning the inventory needed to take the vaults' quote at those pools, the attacker set the vault's internal accounting to a state where it would send collateral into the pools as part of normal option-writing logic. Once collateral was parked, the pools' liquidity accounting let the attacker withdraw it under their own LP position — the flash loan is repaid, the vault is short the collateral, and the attacker keeps the difference.

This is the "collateral enters a pool the attacker controls, then leaves as LP withdrawal" primitive. It has shown up before in options and perp protocols — the specific control surface here is pool creation + quote inclusion of unverified pools. If the quote path accepted only whitelisted pools, the attacker's self-deployed pools would not have mattered.

Numbers block

- Loss (The Defiant / Phemex)       : ~$437,000
- Loss (Blockaid initial estimate)  : ~$400,500 (still ongoing)
- Vaults affected                   : 6
- Draining transactions             : 2
- Setup transactions                : 1 (deployed 6 option pools)
- Chain                             : Arbitrum
- Collateral assets (headline)      : WBTC, WETH, ARB (vault assets)
- Attacker address                  : not disclosed in public reports
- Status                            : attack reported as ongoing at Blockaid's first post
- Premia response                   : none published at time of writing

Figures cross-checked against The Defiant, Phemex News (which reconstructs the Blockaid-sourced details) and ChainCatcher's relay. The exact pool strikes, drain tx hashes and attacker address are still absent from the public record; expect those to land with Premia's post-mortem.

Impact — what matters for users and vault LPs

The direct user impact is confined to LPs of Premia's underwriter vaults on Arbitrum. The protocol's option traders are not drained by this transaction class; the hole sits on the vault side. A rough bound on the per-LP exposure: $437k / 6 vaults = ~$73k per vault, but the distribution between vaults is unknown and the loss is a share, not a flat per-position haircut.

The second-order impact is a signal to builders of similar vault-plus-pool systems: any protocol that lets anyone deploy a pool and that routes vault accounting through those pools has to decide whether unverified pools are quoted. The specific control to look at:

  • A permissioned pool factory (only governance can list a pool).
  • A quote-time filter (pools must be older than N blocks or have external TVL).
  • A sanity bound on how much collateral any single pool can draw from a vault within a block.

Action checklist

  1. Premia vault LPs on Arbitrum: wait for Premia's post-mortem before adding or removing; the vaults are currently drained, and withdrawal fairness depends on the post-mortem's accounting treatment.
  2. Protocol teams running vault-plus-pool AMMs (option, perp, bond): audit whether your quote path trusts user-deployed pools. If so, impose a whitelist or a block-age filter today.
  3. Analysts: watch for Blockaid's follow-up with the attacker address and tx hashes, and for Premia's root-cause write-up on which contract function was the pivot.
  4. Front-end traders: do not sign Premia vault approvals until the protocol publishes a resumed-operation notice.

Context — the Nth "pool-create + flash-loan + quote-at-pool" drain

The pattern is well-trodden. Prior public cases in the same family include Vyper reentrancy against Curve's WETH pools (2023, $73M+ across multiple pools), Euler Finance donation attack (2023, $197M), and smaller option-vault incidents that never made the headline. The Premia incident is small by dollar count — six figures, not nine — but the primitive is textbook, and the ongoing status at the time of Blockaid's first post means the final loss may rise as later transactions settle. Expect Blockaid's forensic thread and Premia's post-mortem to be the primary records to cite.

Related stories