BlackHart
Go live
Hacks feed/Aquifer

Aquifer

August 31, 2026·Solana·Logic Error
$2.5M
total loss
StatusConfirmed

Why it happened

Attack class: Logic error

The swap trusted a caller-supplied token program, so a fake program could report a payment that never happened.

What would have stopped it

  • Pin the token program to the canonical SPL Token Program
  • Check the input balance after the transfer

Not scored before the exploit

BlackHart had not scored Aquifer before this exploit.

On August 31, 2026, about $2.47 million was drained from Aquifer, a Solana AMM. Per third-party analysis, the swap let the caller choose the token program that handled the input payment, so a fake program reported a payment that never happened and Aquifer paid out real tokens. Aquifer offered an on-chain 20% bounty for return of at least 80%, and nothing had been returned as reported in early September. The team's first statements pointed to compromised wallet access, and it had not published a post-mortem.

User walletssafe(no reports of user wallets affected)
Aquifer program codesafe(last deployed weeks earlier, unchanged, per SonarX)
Aquifer vaultsdrained(about $2.47 million)
Exploit anatomy

Aquifer's swap accepted a caller-supplied token program, and the attacker's program reported success for the input transfer without moving tokens. Aquifer then paid out from its vaults, about 212 times in roughly 45 minutes, to the attacker Solana wallet, and the attacker Ethereum wallet also received funds, as named in Aquifer's on-chain message per Shattered.

The attack in 3 steps. Step 1, Root control: Aquifer swap instruction (source). Step 2, Drain: Attacker wallet (Solana) (attacker), about $2.47M reached this step. Step 3, Exit: Attacker wallet (Ethereum) (exit).

  1. Step 1: Root control

    • Source
      Aquifer swap instruction
      accepts a caller-supplied token program
      trusted its success reply
  2. Step 2: Drain

  3. Step 3: Exit

    • Exit
      Attacker wallet (Ethereum)
      funds swapped and bridged
      none returned as reported
      0x2Dfe9e96...922746

Why each step worked

    • Token program not pinnedcaller chose the program that handled the input transfer
    • No input balance checkmissing controlpayout made without confirming payment arrived
    • Real output paid for fake inputvault tokens left with nothing received
    • No outflow limitsmissing controlvaults paid out in many repeated swaps
Full forensic detail

Step-by-step reconstruction, root cause, counterfactuals, remediation, and disclosure timeline.

Exploit anatomy

1.
The attacker called Aquifer's swap with a token program of their own choosing instead of the canonical SPL Token Program, per DeFiLlama and SonarX.
2.
The attacker's program reported success for the input transfer without moving any input tokens.
3.
Aquifer did not verify that the input balance had increased, so it paid out real tokens from its vaults, about 212 times over roughly 45 minutes per SonarX.
4.
The proceeds, about $2.47 million, went to attacker wallets on Solana and Ethereum, and Aquifer posted an on-chain offer of a 20% bounty for return of at least 80%.

Root cause

Aquifer's swap accepted the token program for the input side of a trade from the caller instead of requiring the canonical SPL Token Program, per DeFiLlama and the SonarX analysis. A malicious program in that slot accepted the transfer instruction and returned success without moving any tokens. Aquifer then released real output tokens from its vaults without checking that the input balance had actually grown. The attacker needed no stolen key. The team's first statements pointed to compromised wallet access as a likely cause and said no contract bug was confirmed, and the team had not published a post-mortem when the sources used here were written, so the swap-validation account comes from third-party analysis of on-chain data. SonarX notes that the defect is inferred from on-chain data and not confirmed against unpublished source code.

Prevention analysis

Pin the token program to the canonical SPL Token Program

Rejecting any other program in the token program slot would have stopped a fake program from reporting payment.

Check the input balance after the transfer

Confirming that the vault's input balance grew by the expected amount before paying out would have caught the missing payment.

Per-period vault outflow limits

Capping vault outflow per period would have limited the drain to a small share of the roughly 45 minute run.

Similar incidents

Raydium legacy pools (2026)

A Solana AMM program released real vault tokens after trusting an input the caller supplied, without checking what had been paid in.

Crypto DAO Pro (2026)

A vault trusted a caller-controlled input where it needed a fixed, checked one.

Remediation

1.Aquifer published an on-chain message offering a 20% bounty for return of at least 80% of the funds by September 3, 2026, per press coverage.
2.Other Solana programs should pin the token program account to the canonical program and confirm balance changes after every transfer they depend on.

Timeline

2026-08-31First unpaid settlement, per SonarX's indexed data.
2026-08-31Last unpaid settlement, about 45 minutes after the first.
2026-09-01Aquifer's upgrade authority publishes an on-chain whitehat offer with a September 3 deadline.
SonarX. Defimon Alerts. BlockSec
Continuous adversarial monitoring

Get your protocol scored across 12 dimensions, or request ongoing coverage.