BlackHart
Go live
Hacks feed/Balancer V1

Balancer V1

September 6, 2026·Ethereum·Integer Overflow
$250K
total loss
StatusConfirmed
View the Balancer V1 score →

Why it happened

Attack class: Precision loss / rounding

The same single-asset join rounding flaw, used again by a different attacker after it became public.

What would have stopped it

  • Wind down deprecated pools quickly
  • Round in the pool's favour

What the score saw

849Damascus, scored May 15, 2026

Our pre-hack assessment of Balancer covered the current versions and counted the age of the first version as a strength. It did not examine the first version's single-asset join math, so this old attack surface was not assessed.

Score dimensions

  • Maturitymissed

A second attacker took about $250,000 from legacy Balancer V1 pools on Ethereum on September 6, six days after the first attack was made public. They used the same rounding flaw in the single-asset deposit math on pools with a nearly empty reserve. Per TokenPost, the whole episode reached about 120 pools and about $1.39 million. Anyone still in a V1 pool should withdraw.

Current Balancer versionssafe(not named as affected in the sources)
Legacy V1 poolsdrained(about $250,000 in this wave)
Whole episodedrained(about 120 pools and about $1.39 million per TokenPost)
Exploit anatomy

A different attacker repeated the rounding attack on legacy V1 pools, with one representative transaction at tx.

The attack in 2 steps. Step 1, Root control: Legacy V1 pools (source). Step 2, Drain: Second attacker (exit), about $250K reached this step.

  1. Step 1: Root control

    • Source
      Legacy V1 pools
      near-empty reserve
      ~ $250K this wave
  2. Step 2: Drain

    • Exit
      Second attacker
      different attacker from the first
      ~ $250K

Why each step worked

    • Rounding in single-asset join mathsame flaw as the first attack
    • Flaw public for six daysa second attacker copied the approach
    • Deprecated pools still openmissing controlliquidity was not moved out after the first attack
Full forensic detail

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

Exploit anatomy

1.
Six days after the first attack became public, a different attacker repeated the same class of rounding attack on other legacy V1 pools.
2.
The attacker used the single-asset join, where rounding paid out more than was deposited when a pool's reserve was nearly empty.
3.
Per TokenPost, the whole episode touched about 120 pools and about $1.39 million in total; Blockaid puts this second wave at about $250,000.

Root cause

This is the same flaw as the first Balancer V1 incident: rounding in the single-asset join math paid out more than was deposited on pools with a nearly empty reserve. The first attack was public on August 31, and a different attacker copied the approach on September 6. The pools were deprecated but still open and still held value, so publication of the flaw gave a second attacker what they needed. Per TokenPost, the whole episode reached about 120 pools and about $1.39 million.

Prevention analysis

Wind down deprecated pools quickly

Draining or closing old pools right after the first attack would have given a copycat nothing to take.

Round in the pool's favour

Rounding that always favours the pool would have made the repeat attack unprofitable.

Direct warning to liquidity providers

A clear, early exit notice reaching all V1 depositors would have shrunk the pool of funds at risk.

Similar incidents

Balancer V1 (first attack, 2026)

The original attack on five legacy V1 pools, using the same rounding flaw.

Remediation

1.Balancer told V1 liquidity providers to exit, per The Defiant.
2.Liquidity providers in any V1 pool should withdraw now.
3.AMM teams should treat a published flaw in one deprecated contract as a flaw in every copy of it.

Timeline

2026-08-31First attack on legacy V1 pools becomes public.
2026-09-06Representative copycat transaction from a different attacker.
Blockaid. SlowMist
Continuous adversarial monitoring

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