Worked example · 2026

CoW Swap / Aave interface $50M USDT→AAVE swap — extreme price impact and MEV backrun — 2026-03-12

Loss

an unnamed wallet swapped ~$50.4M of USDT (held as aEthUSDT) for ~324–327 AAVE (~$36K) through the Aave interface, which routed via CoW Swap, on 2026-03-12. The swap executed as signed: it ran through a ~$73–74K SushiSwap AAVE/WETH pool (331.63 AAVE and 17.65 WETH), into which ~17,957 WETH was pushed (~1,017x the pool's WETH reserve). The user was not an "institution" in any source; rekt describes an anonymous whale on a phone who confirmed the trade after the Aave interface displayed an extraordinary-slippage warning (99% price impact) and required a confirmation checkbox. It was not an exploit. Downstream MEV: a ~$12.5M backrun (rekt: a backrun, not a sandwich; the bot kept ~4,824 ETH, ~$11.1M) and ~$30–34M of ETH paid to Titan Builder as builder revenue. The earlier "MEV extracted ~$50M" figure is wrong. Aave pledged to return the fees it collected on the trade ($110,368 per rekt; Decrypt's headline cites ~$600K).

OAK Techniques observed

OAK-T5.004 (Sandwich / MEV Extraction — the "dark forest" extraction pattern: a large trade routed through thin liquidity creates a price dislocation that MEV searchers compete to arbitrage. The searchers extracted the rebalancing profit while the user bore the price impact. The incident is the canonical "liquidity-fragmentation MEV" sub-class: the trade's extreme price impact, not an attacker's sandwich, created the MEV opportunity, but the extraction mechanism is a backrun, structurally adjacent to T5.004.) OAK-T17.001 (Cross-Venue Arbitrage Price Distortion — the price dislocation between the CoWSwap execution venue (where the trade cleared at extreme slippage) and the broader market (where the asset's true price prevailed) was arbitraged by MEV searchers, who bought the dislocated asset on CoWSwap and sold it on external venues at the true price.)

Attribution

unattributed — the swap executed as the user signed it; there was no exploit or attacker. The MEV searchers who extracted the arbitrage were acting rationally within Ethereum's MEV market structure — they observed a price dislocation and arbitraged it, as the MEV market incentivises them to do. The failure was a missing price-impact guard: rekt notes the previous Aave frontend had a ~30% hard slippage cap that was not carried into the CoW-based integration (not confirmed as deliberate), and Aave later announced "Aave Shield", a default 25% price-impact block.

Key teaching point

The $50M USDT→AAVE swap is the canonical "liquidity-fragmentation MEV" anchor: a large trade routed through inadequate liquidity creates a price dislocation that the MEV market extracts, even without a sandwich attacker or a front-running bot. The user's loss was not from a malicious adversary but from an order whose route ended in a ~$73K pool with no effective price-impact limit (99% impact; 1.21% default slippage setting does not capture it). The structural lesson is that an aggregator route that ends in a thin pool converts the trade's price impact into builder revenue and MEV, regardless of intent. No malicious transaction, sandwich or oracle manipulation was involved. The mitigation is an enforced price-impact limit at the interface and solver layer, not only a slippage setting.

Corrected 2026-10-11

the trade was a wallet swap on 2026-03-12 (~$50.4M USDT into a ~$73K SushiSwap pool for ~324–327 AAVE) executed as signed after a 99%-impact warning, not an "institutional" solver-routing anomaly; MEV was a ~$12.5M backrun plus ~$30–34M Titan builder revenue (not ~$50M); the CoW solver won two auctions but failed to land; Aave pledged a fee refund.