Worked example · 2026-07

BonkDAO — an attacker buys just over 1% of supply on KYC exchanges, meets a low quorum in a 2.9%-turnout ballot, and votes the treasury to itself — BonkDAO / Realms (Solana) — 2026-07-06

Loss: ~$20M realised — roughly 4.426 trillion BONK transferred out of the BonkDAO treasury by the automatic execution of a passed governance proposal. The attacker's cost of acquisition was about $4.4M of BONK bought on the open market (some reporting rounds this to ~$4M). BONK fell about 8% on the news. Upbit, Bithumb, and Kraken suspended BONK deposits and withdrawals. No code was broken and no contract was exploited — the governance system executed exactly as designed.
OAK Techniques observed: OAK-T16.002 (Hostile-Vote Treasury Drain — primary, confirmed mechanism. The attacker acquired voting power by legitimate open-market purchase, submitted a treasury-affecting proposal, voted it through on its own weight, and took the treasury egress. The position was held across the entire proposal lifecycle — submission on 2026-06-30, accumulation on 2026-07-04/05, passage and execution on 2026-07-06 — rather than being assembled and unwound inside a single flash-loan-bracketed block, which is what separates this from T16.001. See techniques/T16.002-hostile-vote-treasury-drain.md). OAK-T9.003 (Governance Attack — the generic parent class, preserved on every T16.002 worked example). Distinctively, no OAK-T9.001 step precedes this one: where the Mango Markets anchor used oracle inflation to manufacture the voting position, this attacker simply bought it at full price on Bybit and Binance, which makes the case the clean market-purchase variant of the class.
Attribution: pseudonymous. No named individual or group. The proposal was submitted on 2026-06-30 by an anonymous wallet nominating itself as the destination for the treasury; a separate wallet then bought just over 1% of BONK supply on Bybit and Binance over 2026-07-04 and 2026-07-05. BonkDAO stated it had identified the exchange wallets used for the purchase and was working with exchanges, bridges, and the Solana Foundation. The purchase leg ran through KYC venues, which creates a real identification path — the attacker spent $4.4M through accounts that at least nominally carry identity documentation — but nothing has been made public. Nothing links the operator to a tracked OAK actor.
Key teaching point: A treasury protected only by token-weighted voting is for sale, and its price is the quorum. BonkDAO's quorum was low enough that just over 1% of supply — $4.4M — was decisive, against a $20M treasury. That is not an exploit; it is an arbitrage the governance parameters advertised openly, and the attacker paid retail on public exchanges to take it. The load-bearing failures are all parameter and process choices, not code. First, quorum was set far below the value it guarded, so the treasury could be bought at a 4.5x discount. Second, there was no execution timelock: passage and treasury egress were the same event, leaving no window in which 18,000 non-voting holders could react to a proposal that had a line reading "all YES voters are eligible to receive tokens" in its own pitch. Third, nobody was watching the accumulation — a wallet buying 1% of supply on centralised venues across two days, six days after an anonymous proposal to move the treasury to a wallet it controlled, is a correlation with a very small candidate set. The controls that close this are unglamorous governance-process ones: quorum scaled to treasury value, a mandatory timelock between passage and execution long enough for a counter-proposal or emergency veto, and surveillance of voting-weight accumulation against open proposals.

Summary

BonkDAO is the community governance body for BONK, a Solana memecoin. Its treasury was governed through Realms, Solana's token-weighted on-chain governance platform, in which voting power is proportional to tokens held and a passed proposal's payload executes automatically.

On 2026-06-30, an anonymous wallet submitted "BIP #76 — Sowellian BonkDAO", a proposal whose payload transferred the DAO treasury's BONK holdings to a wallet the proposer controlled. Its written pitch made little attempt at plausibility, promising to "rebuild from the ashes, monetize holdings, stop the bleeding" and noting that "all YES voters are eligible to receive tokens."

Over 2026-07-04 and 2026-07-05, a separate wallet spent roughly $4.4M buying just over 1% of BONK's circulating supply on Bybit and Binance — enough to meet BonkDAO's quorum threshold on its own.

On 2026-07-06, BIP #76 passed. Seven wallets voted. More than 18,000 eligible holders did not, producing a turnout of about 2.9%, and the proposal carried with roughly 99.9% "yes". The purchased position was decisive by itself. Execution was automatic and immediate: approximately 4.426 trillion BONK, worth about $20M, left the treasury.

BonkDAO confirmed the attack, described it as a malicious governance proposal, notified law enforcement, and said it had identified the exchange wallets used to acquire the tokens ahead of the vote. Upbit, Bithumb, and Kraken suspended BONK deposits and withdrawals. No recovery has been confirmed. The community's response has centred on governance reform — execution timelocks, higher quorums, and emergency multisig controls — none of which existed at the time.

Timeline (UTC)

When	Event	OAK ref
(standing)	BonkDAO treasury governed by token-weighted Realms voting: quorum reachable with ~1% of supply, no execution timelock, no emergency veto	(standing T16.002 surface)
2026-06-30	Anonymous wallet submits "BIP #76 — Sowellian BonkDAO", payload transferring the treasury to a wallet it controls; pitch states "all YES voters are eligible to receive tokens"	T16.002 setup
2026-07-04 – 07-05	A separate wallet buys just over 1% of BONK supply on Bybit and Binance for ~$4.4M — enough to meet quorum alone	T16.002 acquisition
2026-07-06	BIP #76 passes: 7 wallets vote, 18,000+ eligible holders do not, ~2.9% turnout, ~99.9% "yes"	T16.002 passage
2026-07-06	Payload executes automatically and immediately; ~~4.426T BONK (~~$20M) leaves the treasury	T16.002 extraction
2026-07-06 onward	BONK falls ~8%; BonkDAO confirms the attack, notifies law enforcement, identifies the exchange wallets, coordinates with exchanges, bridges, and the Solana Foundation; Upbit, Bithumb, and Kraken suspend BONK deposits/withdrawals	(operator response)

What defenders observed

Pre-event (quorum priced below the treasury it guards). The single decisive parameter: ~1% of supply ($4.4M) carried a $20M treasury. Quorum should be set against the value at risk, not as a fixed token fraction chosen at launch and never revisited as the treasury grew. The pre-event question every DAO can answer today is what does it currently cost to pass a proposal here, and is that number smaller than what we hold? (M17).
Pre-event (no timelock: passage and payout were one event). With no delay between a proposal carrying and its payload executing, the 18,000 holders who had not voted had no window in which to notice, organise, or veto. A timelock is what converts a governance attack from a fait accompli into a race the defenders can enter. The absence of an emergency multisig with veto scope compounded it (M17, M34).
Pre-event (2.9% turnout is the exploitable condition, not a side note). Token-weighted governance is only as strong as participation; a treasury guarded by a quorum that habitual non-participation leaves within reach of any well-funded buyer is a standing offer. Chronic low turnout should be read by risk teams as a live governance-capture exposure with a computable price.
At-event (voting-weight accumulation against an open proposal). The highest-leverage detection signal here needs no on-chain novelty: a wallet acquiring ~1% of supply across two days, while an anonymous proposal to move the treasury sits open, is a correlation with a tiny candidate set. Cohort surveillance of accumulation against open proposals — especially proposals whose beneficiary overlaps the proposing cohort — is the T16.002 detection control, and BonkDAO had no consumer for a signal that was fully visible in advance (M04, M05).
At-event (the proposal disclosed its own intent). BIP #76's text said YES voters were eligible to receive tokens. Proposal-text screening is crude but nearly free, and this proposal would have failed any of it. The reason it passed is that almost nobody read it — which is the turnout problem restated.
Response (KYC purchase is the residual identification path). The attacker's $4.4M ran through Bybit and Binance accounts. That is the strongest recovery and attribution lever available after the fact, and it is the reason BonkDAO's coordination with exchanges is more than symbolic. It is also the one part of the operation the attacker could not perform pseudonymously.

Public references

[coindeskbonkdao2026] — CoinDesk, "BONK faces $20 million treasury drain after attacker spends $4 million to pass malicious proposal" (loss, acquisition cost, malicious-proposal framing, exchange coordination): https://www.coindesk.com/markets/2026/07/07/bonk-faces-usd20-million-treasury-drain-after-attacker-spends-usd4-million-to-pass-malicious-proposal
[techtimesbonkdao2026] — Tech Times, "BonkDAO Loses $20M as Attacker Buys Quorum With $4.4M in BONK" (the $4.4M / just-over-1%-of-supply quorum purchase, Bybit and Binance venues, 2026-07-04/05 accumulation window, 99.9% yes): https://www.techtimes.com/articles/319820/20260707/bonkdao-loses-20m-attacker-buys-quorum-44m-bonk.htm
[cryptonewsbonkdao2026] — crypto.news, "The BONK governance attack: How a DAO lost $20 million in one proposal" (BIP #76 title and pitch text, 2026-06-30 submission by an anonymous wallet, 7 wallets voting against 18,000+ non-voters, 2.9% turnout): https://crypto.news/the-bonk-governance-attack-how-a-dao-lost-20-million-in-one-proposal/
[cryptobriefingbonkdao2026] — Crypto Briefing, "BonkDAO treasury drained of $20 million in BONK after malicious proposal" (DAO confirmation, law-enforcement notification, Solana Foundation coordination): https://cryptobriefing.com/bonkdao-treasury-drained-20m-malicious-proposal/
[cryptotimesbonkdao2026] — The Crypto Times, "Crypto Loses $35M in a Week: BonkDAO, Bonzo Lend, Summer.fi Hacked" (~4.426T BONK figure, Realms platform, exchange suspensions at Upbit / Bithumb / Kraken, governance-reform demands): https://www.cryptotimes.io/2026/07/12/crypto-loses-35m-in-a-week-bonkdao-bonzo-lend-summer-fi-hacked/

Discussion

BonkDAO is the variant OAK's T16.002 page has been missing: the class without an upstream T9.001. The Mango Markets anchor (October 2022) needed oracle manipulation to inflate an MNGO position into decisive voting weight, and that dependency has quietly shaped how the class is read — as the second act of a more sophisticated exploit. Here there is no first act. The attacker walked into Bybit and Binance, paid full market price for just over 1% of supply, and that was sufficient. Stripping out the oracle step exposes the class's actual premise, which is simply that token-weighted governance sells control at the quorum price, and that the price is frequently a large discount to the treasury.

The comparison worth drawing inside OAK is with examples/2026-06-token-of-power-top-low-float-governance-takeover-mint.md, where a low float made governance capture cheap. BONK is not a low-float token — it is a widely-distributed memecoin with 18,000+ eligible governance participants — and it was captured anyway, because distribution without participation is indistinguishable from concentration at the ballot. A defender modelling governance-capture risk from holder distribution alone would have scored BonkDAO as safe. The metric that mattered was turnout, and 2.9% turnout means an attacker is not competing against 18,000 holders but against the handful who actually vote.

The uncomfortable part of this case is how much of it was visible in advance and how little of that visibility was wired to anything. The proposal was public from 2026-06-30 and openly promised tokens to YES voters. The accumulation was public, on centralised venues, across two days, in a size that had no other plausible explanation. The quorum arithmetic was a matter of published parameters. Every element was observable, and the gap was purely that no one had built the consumer — no cohort surveillance of accumulation against open proposals, no timelock to give a reaction window, no emergency veto to use if anyone had reacted. Contributors documenting future governance captures should record the quorum-cost-to-treasury-value ratio explicitly; it is the single number that most cleanly predicts this class, and it is knowable before the attack rather than after.