BlackHart
Go live
Hacks feed/Term Finance (Term Labs vaults)

Term Finance (Term Labs vaults)

August 23, 2026·Ethereum·Governance Attack
$8.5M
total loss
StatusConfirmed

Why it happened

Attack class: Governance attack

Control of the vaults ran through a thin wrapped-share vote, and the vote could remove the Zodiac Delay cooldown that protected depositors.

What would have stopped it

  • Governance cannot change its own delay
  • Voting weight tied to real stake

Not scored before the exploit

BlackHart had not scored Term Finance (Term Labs vaults) before this exploit.

An attacker took about $8.5 million from the Term Labs vaults on Ethereum, made up of 2,841.74 WETH and 1,679,639.29 USDC. Voting on the vaults required wrapped vault shares and almost none existed, so the attacker gained a majority vote cheaply, passed proposals that removed the waiting period on withdrawals, and moved the money out. Depositors in the ETH Meta Vault and five USDC vaults were affected. About 960 ETH had reached Tornado Cash by September 4 and no funds were returned.

Vaults outside the affected setsafe(not named in the post-mortems as hit)
ETH Meta Vaultdrained(2,841.74 WETH drained)
Five USDC vaultsdrained(1,679,639.29 USDC drained)
Exploit anatomy

The attacker gained a majority vote over the Term Labs vaults, passed proposals that removed the withdrawal delay, and drained the ETH vault in tx and the USDC vaults in tx.

The attack in 3 steps. Step 1, Root control: Term Labs vaults (source). Step 2, Drain: Attacker (attacker), about $8.47M reached this step. Step 3, Mixer: Tornado Cash (exit), about $3M reached this step.

  1. Step 1: Root control

    • Source
      Term Labs vaults
      ETH Meta Vault plus five USDC vaults
      ~ $8.5M total
  2. Step 2: Drain

    • Attacker
      Attacker
      2,841.74 WETH
      1,679,639.29 USDC
  3. Step 3: Mixer

    • Exit
      Tornado Cash
      about 960 ETH in by September 4
      no funds returned

Why each step worked

    • Voting needed wrapped sharesalmost none were wrapped, so a majority was cheap
    • Delay removable by the same voteZodiac Delay cooldown taken out by proposal
    • No quorum floor or outflow capmissing controlone passed proposal opened all vaults
    • No recoverymissing controlfunds were not returned
Full forensic detail

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

Exploit anatomy

1.
Voting power over the Term Labs vaults required wrapped vault shares, and almost none had been wrapped, so the attacker was able to gain a majority vote over the ETH Meta Vault and five USDC vaults cheaply.
2.
Using that majority, the attacker passed governance proposals on the vaults that removed the Zodiac Delay cooldown, the waiting period that normally gives depositors time to react.
3.
With the cooldown gone, the attacker moved the ETH vault funds out in one transaction, taking 2,841.74 WETH.
4.
The same route was used on the five USDC vaults about 22 minutes later, taking 1,679,639.29 USDC.
5.
Term acknowledged the incident at 07:32:40Z. PeckShield later reported about 960 ETH moved into Tornado Cash by September 4. No funds were returned.

Root cause

The Term Labs vaults let holders of wrapped vault shares vote on proposals that controlled the vaults themselves. Per the Defimon and Rekt write-ups, almost nobody had wrapped any shares, so the voting supply was tiny compared with the money the vaults held. That meant a majority vote could be gained cheaply, far below the value it controlled. The attacker used that majority to pass proposals that removed the Zodiac Delay cooldown, the control meant to give depositors time to exit or the team time to intervene, and then moved the funds in ordinary transactions. The design flaw was tying control of the whole vault to a thin, easily acquired voting base while the one delay that could have caught the attack was itself changeable by the same vote.

Prevention analysis

Governance cannot change its own delay

A cooldown that the same vote can remove gives no protection. Making the delay immutable, or requiring a separate longer process to change it, would have left depositors time to withdraw.

Voting weight tied to real stake

Counting all vault shares, wrapped or not, would have made a majority cost a large share of the vault rather than a trivial amount.

Quorum floor and outflow limits

A minimum participation level and a cap on how much can leave per period would have limited the loss from a single passed proposal.

Similar incidents

X100 (2026)

Control of the treasury went to whoever held enough voting weight at the moment of the vote, with no delay between winning control and using it.

Remediation

1.Term acknowledged the incident publicly and the affected vaults were halted, per its announcement.
2.Vault protocols that gate control on an opt-in wrapper token should check how much of the underlying is actually wrapped, and base voting on the full share supply.
3.Users should withdraw from any vault where governance can remove its own timelock.

Timeline

2026-08-23ETH vault drain transaction.
2026-08-23USDC vault drain transaction.
2026-08-23Term Labs acknowledges the incident.
2026-09-04PeckShield reports about 960 ETH deposited into Tornado Cash.
Defimon. Rekt. PeckShield
Continuous adversarial monitoring

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