Balancer DAO

Balancer DAO governs Balancer, a decentralized exchange built as a programmable automated market maker: instead of the fixed 50/50 pools of a Uniswap, Balancer pools can hold up to eight assets in arbitrary weights, which makes them double as self-rebalancing index funds. Its Vault architecture – one contract that holds all pool assets and nets transfers across them – became one of the most-forked designs in DeFi. But Balancer's real significance to DAO governance is its adoption, and eventual abandonment, of the vote-escrow (ve) model: it is a case study of a flagship protocol locking its token for governance, running that system for four years, and then unwinding it in the open. In September 2026 its holders voted to wind the protocol itself down and pay the treasury out to themselves. (BAL token docs)

veBAL – the 80/20 vote-escrow lock

BAL launched in June 2020 with a fixed 100M maximum supply, but liquid BAL never carried governance weight. In March 2022 Balancer adopted the veModel, a direct descendant of Curve's veCRV, with one distinctive twist: you did not lock BAL itself. You locked an 80/20 BAL/WETH Balancer Pool Token (BPT) – a liquidity position that is itself 80% BAL – for up to one year, and received non-transferable veBAL whose weight scaled with lock length and decayed toward the unlock date. Locking the LP token rather than the bare asset meant governance participants were also providing protocol liquidity, not just idling tokens in an escrow.

veBAL bundled the same three rights the ve-model everywhere offers: a vote on protocol proposals, a share of protocol fees, and – the load-bearing one – a weekly vote on gauge weights that decided how BAL emissions were split across pools. As with Curve, that gauge vote created a market: Aura Finance rose as an aggregation layer (Balancer's Convex), accumulating veBAL and renting out its emission-steering power to projects that wanted deep liquidity. The vote-market dynamic the Curve Wars made famous replayed on Balancer.

November 2025 – the $128M v2 exploit

On 3 November 2025 Balancer suffered one of the largest DeFi exploits of the year: an attacker drained roughly $128 million from Balancer v2 pools across six chains – Ethereum, Base, Polygon, Arbitrum and others – in under 30 minutes. The bug was not a stolen key or a governance takeover but an arithmetic precision error: a rounding flaw in the Vault's _upscaleArray scaling logic, which – chained through dozens of tiny crafted batchSwap operations executed inside a contract constructor – let the attacker suppress BPT prices and repeatedly extract value. It is a textbook entry for the DAO security and governance-attacks page: the DAO's votes worked fine; the loss came from a subtle invariant-math bug in audited, years-old code. (Check Point Research analysis)

2026 – Balancer Labs winds down, and so does veBAL

The exploit's fallout reshaped the whole project. In March 2026 the founding company, Balancer Labs, announced it was dissolving its corporate entity – a co-founder wrote that after the exploit the entity "became a liability" – while the protocol itself stayed live under DAO stewardship. That forced the DAO to confront a question every maturing DAO eventually faces: is the emission-subsidised ve-machine still worth its overhead once the growth phase is over?

Its answer, ratified in Q2 2026, was to dismantle it:

• BIP-919 (BAL Tokenomics Revamp) – end all new BAL emissions, lift the LP share of swap fees from 50% to 75%, and route 100% of protocol fees to the DAO treasury. With no emissions left, the gauge system had nothing to steer.
• BIP-920 (veBAL Compensation Airdrop) – a 500,000 USDC airdrop to veBAL lockers, compensation for locked positions that would no longer earn incentives.
• BIP-921 (1-BAL-1-vote) – replace the two veBAL Snapshot strategies with a seven-strategy stack counting raw BAL across every production chain: no lockup, no decay. Governance reverted to plain token-weighted voting.

The arc is the lesson. The ve-lock was adopted to bind control to long-term commitment; four years on, with emissions gone and the vote-market's overhead exposed, Balancer judged the lock no longer paid for itself and returned to the one-token-one-vote model the lock was invented to escape.

Balancer was not alone. Months later Pendle retired its own vePENDLE gauge system for a liquid staking token (sPENDLE) plus algorithmic emissions – the same finding, that the ve-lock concentrated power in a rentable few while most holders sat out, reached independently by another of the model's biggest adopters.

August 2026 – the aggregator is evicted, and the quorum goes with it

The last act of the ve-model is not the end of veBAL but the end of the thing that grew on top of it. Read from the balancer.eth Snapshot space on 22 August 2026:

• Aura wound itself down first. Once Balancer sunset veBAL there was nothing left for a vote-aggregator to aggregate. AIP-78, Orderly Protocol Wind-Down and Treasury Redemption for Aura ran 9–12 April 2026 in Aura's own space and closed on eight voting addresses carrying 17,171,689 in favour, zero against, zero abstaining. The locker was shut and the meta-governance layer with it.
• Balancer is now formally excluding it. BIP-924, filed 14 August 2026, passed on 18 August on 23,891,200.51 BAL in favour, none against, from seven voting addresses – a tally equal to 33.0% of BAL’s 72,373,419.35 total supply at the proposal’s own snapshot block 25,754,324. It overwrites the voting power of the Aura VoterProxy at 0xaF52695E…0Dbec2 to zero in the balancer.eth aggregator. Aura's underlying BAL becomes eligible again only as individual users redeem it, after an unlock dated 29 April 2027.
• And it has to cut quorum to keep governing. The proposal states the reason plainly: Aura was responsible for "at least 60% (usually >75%) of the votes of every Snapshot proposal," and although BIP-521 capped its voting power, all of it still counted toward quorum. The 10m BAL bar was calibrated around a participant that no longer exists, so BIP-924 halves it to 5m BAL.

The two proposals immediately before it show why that is not housekeeping. BIP-922 and BIP-923 both closed on 4 August 2026 with every vote in favour and none against – and with 11 and 10 voting addresses carrying 6.47m and 6.45m BAL, against a 10m quorum. A DAO that outsourced three-quarters of its turnout to an aggregator discovers, when the aggregator leaves, that its own rules were written for a body it no longer has.

There is a closing detail worth keeping. Of the 500,000 USDC that BIP-920 airdropped to veBAL lockers as compensation, 336,074.19 – just over 67% – went to that same Aura VoterProxy address. The DAO compensated the aggregator for the end of the lock, and four months later moved to zero out its vote.

Three days after the new bar came into force, Balancer put the failed proposals back. On 21 August 2026, between 09:16 and 09:22 UTC, three proposals opened in the space carrying quorum 5,000,000 rather than 10,000,000: BIP-925, Unwind Timeless ve8020 BAL Grant, and explicit reruns of BIP-922 and BIP-923 – the two proposals that closed unanimously on 4 August at 6.47m and 6.45m BAL and failed only the old quorum. All three passed when voting closed on 25 August, each on 6.52m BAL from seven addresses with none against: over the new 5m bar, and still short of the old one. The sequence shows that a quorum is a design parameter and not a fact about consent: the same proposals, the same electorate, the same unanimity, decided differently by a number the DAO chose.

September 2026 – the DAO votes to wind Balancer down

On 29 September 2026 Balancer's BAL holders voted to close the protocol and pay its treasury out to themselves. BIP-928, Orderly Winddown of Balancer and Distribution of the Treasury, passed with 17.09m BAL for across its two Yes options and 139,573 against, from 42 voting addresses. The vote had been reset on 26 September, after a holder found the space's 1-BAL-1-vote strategy was not applying delegate overrides (forum thread). The proposal's case is that the restructuring holders approved in April had not produced the revenue it needed: most protocol revenue still came from v2, and v3 had not grown to replace it.

Its terms set the order of the shutdown:

• Pools that can be paused move to withdrawals only on 30 October 2026, when the bug bounty also ends. v3 pools a partner asked to keep for a migration stay live until the v3 Vault is paused on 30 November.
• The treasury, at least $9m on the reading of kpk, the DAO's treasury manager, when the proposal was posted, is paid out in kind and pro rata. Round one opens at the end of May 2027, after every existing veBAL lock has expired: holders burn BAL for their share over a six-month window. A second round and a final sweep go only to the addresses that redeemed in the first.
• BAL itself is not paid out, and the BIP-919 buyback is cancelled. Holders of tetuBAL, a permanent lock that can never turn back into BAL, receive treasury BAL for 50% of the BAL behind their position, the option that drew more of the Yes vote.
• The wind-down has a budget of $150k from 1 November 2026 to May 2027 and $30k from then to the final sweep, with a $220k reserve drawn only if needed. The DAO's admin and multisig roles are revoked or transferred as each becomes safe to retire.

Three more votes followed from it. BIP-929, from the MAXYZ team, asked for the DAO's remaining non-circulating BAL to seed an official fork, in return for 10% of any future token supply; it failed on the same day, 5.21m BAL for and 12.11m against. Its replacement, BIP-932, asked only for the DAO's blessing to fork and for access to its security material, with no grant and no use of the Balancer name, and passed on 13 October with none against. BIP-931, passed on 6 October, sells the DAO's COW back to CoW DAO's treasury for ETH at a 15-day average price and returns the 200,000 BAL of a January 2025 token swap, so that redeemers are not handed an illiquid governance token.

The oldest code failed during the wind-down. On 31 August 2026 an attacker used a rounding flaw in the single-sided join of Balancer v1, the immutable 2020 pools that cannot be paused, and copycats repeated it across the remaining v1 pools; about $1.39m was drained at the time's prices. BIP-930, passed on 6 October, pays the affected liquidity providers from what whitehats and the main exploiter returned, under the framework used after the November 2025 attack. How a DAO closes in general is set out under winding down a DAO.

How Caper approaches this

Balancer's ve-experiment ran the full loop: lock the token to couple influence to commitment, watch an aggregator tokenise and rent that locked power, then – once emissions stopped justifying the machinery – unwind the whole thing back to plutocratic token-weighted voting. A caper reaches for the same goal the lock was after – influence tied to genuine stake – but without a lock to tokenise or a gauge to fight over. Its voting weight takes in the tokens a member holds but grows with the votes they have earned through participation, so a large position alone cannot capture control and there is no locked receipt for a third party to aggregate and sell.

There are also no emissions: a caper's tokens are minted along a bonding curve, not a weekly inflation vote, so there is no emission stream for a "war" to capture in the first place. And because that same earned weight is each member's exit claim on the treasury, control stays coupled to real economic exposure rather than to whoever locked the most this cycle. This is a design contrast, not a claim of superiority – the mechanics are on the linked pages and verified against the caper contract source.

References

• Balancer docs – BAL token and veBAL (primary).
• Balancer Labs, Unlocking the veModel – the 2022 veBAL launch rationale (primary).
• BIP-919, BIP-920, BIP-921 – the 2026 tokenomics revamp and veBAL deprecation, on the Balancer governance forum (primary).
• Check Point Research, How an attacker drained $128M from Balancer (November 2025).
• CoinDesk, Balancer Labs to shut down after exploit (March 2026).