A $320,000 trade just triggered $36 million in liquidations

On the morning of Aug. 25, 2026, roughly $320,000 changed hands in a thinly traded Ethereum yield token — and within 14 minutes, $36 million of leveraged loans were forcibly closed. Nothing was hacked, and the dollar asset underneath never lost its peg.

What Triggered the $36 Million PT-reUSD Liquidation Cascade?

The cascade was triggered by an 11-trade buying spree in YT-reUSD that dragged the PT-reUSD price feed used by a Morpho lending market down by about 3%. Between 04:37:47 and 04:51:23 UTC on Aug. 25, 2026, 33 liquidation events repaid $36.14 million of debt and seized 38.6 million principal tokens — $35.19 million from the USDC market and $956,000 from the USDT market — with zero realized bad debt on either . No smart contract was exploited and ETH did not crash; the cause was a short oracle averaging window reading a shallow pool .

Quick Answer: A single wallet spent about $320,000 buying YT-reUSD on Pendle, mechanically pushing the paired PT-reUSD price down roughly 3%. That move hit a 15-minute TWAP oracle on Morpho and liquidated $36.14 million of debt across 33 events in 14 minutes — no hack, no depeg.

Blockchain security firm PeckShield flagged wallet 0x854e…690d, which executed 11 consecutive Pendle trades between 04:28 and 04:37 UTC, converting roughly $320,000 of SY-reUSD into more than 9.5 million YT-reUSD before exiting . Because Pendle splits a yield-bearing asset into a principal token and a yield token carved from the same underlying, bidding the yield side up mechanically repriced the principal side lower — implied annual yield jumped from roughly 11% to above 20%, and PT-reUSD fell about 3% .

Re Protocol, the issuer of reUSD, described the event as "a market-price movement in the Pendle PT-reUSD oracle used by a third-party Morpho market" and confirmed reUSD itself was unaffected, per CoinDesk . Pendle stated that the affected PT-reUSD/USDC market was deployed by Steakhouse on Morpho and that the oracle "was set up correctly and functioned as intended," per The Defiant .

The size mismatch explains the leverage in the trigger. The Dec. 10 Pendle reUSD pool held about $8.97 million of liquidity, while the Morpho market pricing off it carried roughly $67.5 million of PT collateral against $52.2 million of borrows . A lending market several times larger than the pool that prices its collateral is a structural exposure, not an accident of that morning.

The Mechanics: How a $320K Yield-Token Trade Crashed a $52M Lending Market

The mechanism was a seesaw, not a hack. Pendle splits a yield-bearing asset into a principal token (PT) and a yield token (YT), both carved from the same underlying reUSD position, so bidding YT up mechanically makes PT cheaper — the market is simply repricing the yield component higher . Between 04:28 and 04:37 UTC on Aug. 25, 2026, wallet 0x854e…690d converted roughly $320,000 of SY-reUSD into more than 9.5 million YT-reUSD across 11 consecutive trades . PT-reUSD fell about 2.8–3% as a direct consequence.

Pendle's documentation describes the PT as a zero-coupon-bond-like claim redeemable for the underlying at maturity — here Dec. 10, 2026 — while the YT captures the variable yield stream until that date . The two prices are bound by that shared origin. Buying yield aggressively in a burst tells the market that the remaining yield stream is worth more, which arithmetically leaves less value in the principal claim. On CoinDesk's reconstruction, the burst pushed the market's implied annual yield from roughly the 11% area to above 20% . That repricing of the yield leg was the entire source of the PT discount — no depeg, no contract failure.

What made a $320,000 order move a market at all was the size mismatch inside the pool. Onchain analyst 0scar put roughly 5.4 million PT-equivalent of flow against a pool holding about 3.1 million PT, which dropped the 15-minute time-weighted average price to 0.9647 . A trade nearly twice the depth of the book it hits does not transact at the quoted price; it walks the curve. In a long-dated pool where most of the principal has already been absorbed by loopers and locked as collateral elsewhere, the float available to absorb that order was thin by construction.

"Investigating whether the PT market price was intentionally manipulated and are working with the relevant teams on a safer oracle configuration," Re Protocol, the issuer of the underlying reUSD asset, said in its post-incident statement, while confirming reUSD itself was unaffected (source: The Defiant).

The scale gap between the pricing venue and the priced market is the part worth internalizing. Per The Defiant's figures, the Dec. 10 Pendle reUSD pool carried about $8.97 million of liquidity, while the Morpho market referencing it held roughly $67.5 million of PT collateral against $52.2 million of borrows . Three ratios follow from those numbers:

• Collateral-to-pool ratio: roughly 7.5x. Every dollar of pool depth was setting the mark for about $7.50 of collateral.
• Borrows-to-pool ratio: roughly 5.8x. Outstanding debt priced off a venue that could not have absorbed an unwind of comparable size.
• Trigger-to-borrow ratio: about 0.6%. The $320,000 order was a fraction of a percent of the $52.2 million it ultimately repriced.

Morpho's own documentation states that each market's oracle is selected at market creation, that the protocol is oracle-agnostic, and that oracle choice defines the market's risk profile . The pricing venue was therefore a deliberate parameter, and its depth relative to the market it fed was the variable that turned a routine yield trade into a systemic input.

Why Leverage Turned a 3% Price Move Into $36M in Forced Selling

A 3% collateral move only liquidates a market when borrowers are already sitting within 3% of their limit — and that is precisely what recursive looping had produced on Morpho's PT-reUSD markets. Traders deposited PT-reUSD-10DEC2026, borrowed USDC or USDT against it, used the borrowed dollars to buy more PT-reUSD, and redeposited — repeating the cycle to multiply a fixed, known yield . Each turn of the loop raised effective exposure and simultaneously shrank the price cushion between the position and its liquidation threshold.

The carry trade itself is rational arithmetic. A principal token accretes toward $1 at maturity on a schedule, so the return is fixed if held to Dec. 10, 2026; borrowing at a lower rate against it converts a modest spread into a larger one. What looping cannot change is the interim path. Before maturity, the position is marked against a live oracle, and leverage compresses the tolerance for any deviation in that mark — including deviations that have nothing to do with the solvency of the underlying dollar asset .

The parameter that set the ceiling on how far loopers could push was the liquidation loan-to-value. PT-reUSD markets paired against USDC, USDT and AUSD on Ethereum all carried a 91.5% LLTV, a figure confirmed in both Steakhouse's Morpho v1 market registry and Morpho's own network data . An LLTV that high is a deliberate statement that the collateral is expected to behave like a dollar. It leaves roughly 8.5 percentage points of theoretical room — but loopers, by design, consume nearly all of it.

How little remained is documented on both sides of the incident. CoinDesk reported that affected borrowers had less than 3% of protection before automatic liquidation triggered . Steakhouse's own account after the event flagged positions with health factors below 1.03 as vulnerable and measured the PT asset move at 2.8% on high volume — a deviation that fits inside a 3% buffer with room to spare .

The tail of that distribution was thinner still. Onchain analyst 0scar identified one borrower carrying 90.9% LTV against the 91.5% threshold . Run through Morpho's published formula — HEALTH_FACTOR = (COLLATERAL_VALUE × LLTV) / BORROWED_AMOUNT — that position sat at a health factor near 1.007, roughly seven-tenths of one percent from forced sale . At that distance, the question is not whether a shock arrives but which shock arrives first.

Three structural features combined to convert a small mark change into $36 million of forced repayment:

• Correlated positioning. Loopers in a PT carry trade are not diversified against one another — they hold the same collateral, borrow the same stablecoins, and cluster at the same LLTV ceiling, so a single oracle print moves the entire cohort across the threshold simultaneously.
• Buffer consumed by design. Leverage yield-farming is optimized to minimize idle collateral; the more efficient the loop, the closer the health factor sits to 1.0 and the smaller the tolerable deviation.
• Modest liquidator incentive, ample supply. Morpho's liquidation incentive factor — LIF = min(1.15, 1/(β×LLTV+(1−β))) with β=0.3 — yields only about a 2.6% bonus at 91.5% LLTV . Thin as that premium is, it was sufficient to clear the entire eligible book within minutes.

The result was a cascade with no code failure at any layer. Every contract executed as specified; the liquidations were valid under the market's own rules. The exposure came from stacking a high-efficiency LLTV on top of leverage that consumed the remaining margin, leaving a cohort of borrowers whose collective survival depended on a single oracle feed never moving more than a few tenths of a percent against them.

Decision Criteria: How to Evaluate Risk Before Looping a PT-Collateral Market

Evaluating a PT-collateral lending market means checking four parameters before depositing anything: the oracle's averaging window, the ratio of pricing-pool liquidity to market borrows, the liquidation loan-to-value relative to time-to-maturity, and the oracle's type. The PT-reUSD market on Morpho failed all four simultaneously — a 15-minute TWAP priced roughly $52.2 million of borrows off an $8.97 million pool at a 91.5% LLTV . Any one of those in isolation is survivable. Composed, they are not.

1. Oracle window length. Pendle's integration guidance names 900 seconds (15 minutes) as typical for most markets and 1800 seconds (30 minutes) for others — but explicitly conditions those defaults on choosing markets with deep liquidity and high trading activity . A 15-minute window is a cost function: it sets how long an attacker must hold a distorted price, and therefore how much capital they must risk. RedStone's dynamic PT oracle uses one- to four-hour windows for precisely this reason, arguing that longer averaging makes manipulation prohibitively expensive . Before looping, find the window. If it is 900 seconds and the pool is shallow, the defense budget is measured in minutes.

2. Pool-to-market ratio. This is the single most diagnostic number and it is publicly checkable. Compare the Pendle pool's liquidity against the lending market's total borrows. On Aug. 25, 2026, the Dec. 10 reUSD pool held about $8.97 million against a Morpho market carrying roughly $67.5 million of PT collateral and $52.2 million of borrows — the lending market was roughly six times the size of the venue setting its price. When that ratio inverts, the cost of moving the oracle falls far below the value of the positions it governs. A useful screen: if total borrows exceed pool liquidity by more than about 2x, the oracle is structurally undersized regardless of how it is configured.

3. LLTV versus time-to-maturity. Morpho's PT-reUSD markets against USDC, USDT and AUSD on Ethereum all carried a 91.5% LLTV . That parameter is defensible near maturity, when a PT converges mechanically toward $1 and secondary-market noise shrinks. It is a different instrument at 15 months out. A Dec. 10, 2026 maturity leaves a long stretch during which the PT's price is driven by rate expectations and pool depth, not by redemption arithmetic. Treat LLTV and maturity as a single joint parameter, not two independent ones — and size leverage against the longer of the two horizons.

4. Oracle type. There is no clean answer here, only a documented tradeoff. Steakhouse mapped it publicly in June 2025: market-based oracles track true price but are thin and manipulable, while hardcoded or curve-based oracles resist manipulation but can miss a genuine depeg . The PT-reUSD oracle used the lower of a 15-minute market TWAP or a fixed accreting curve, so the manipulable leg became operative exactly when it was most dangerous. Chaos Labs has made the structural version of this argument — feeds pulled directly from Pendle AMM data are vulnerable whenever pools become imbalanced or sparse .

"Morpho is oracle-agnostic: each market's oracle is chosen at market creation, and that choice defines the market's risk profile. No oracle is immune to manipulation that can lead to liquidations or bad debt," per Morpho's protocol documentation (source: Morpho Docs — Oracles and Morpho Docs — Risks).

That framing shifts the burden onto the depositor. In a permissionless, isolated-market design, no one upstream is guaranteeing that a given oracle suits a given collateral. The four checks above take about ten minutes and are all verifiable from public pages — the Pendle pool page, the Morpho market page, and the curator's market registry. CoinDesk reported that affected borrowers had less than 3% of protection before automatic liquidation . None of the four parameters that produced that outcome were hidden.

PT-reUSD Morpho Market vs. Safer PT Lending Setups: Side-by-Side

The same 91.5% liquidation LTV can be conservative or reckless depending on three companion parameters: TWAP window, pool-depth-to-market ratio, and oracle type. On the PT-reUSD-10DEC2026 market, a $8.97 million Pendle pool priced roughly $67.5 million of collateral and $52.2 million of borrows , a depth ratio near 0.13x, read through a 15-minute average. Change any one of those inputs and the $320,000 trade stops being enough.

Pendle's own integration guidance names 900 seconds (15 minutes) as typical for most markets and 1800 seconds (30 minutes) for others, but conditions both on selecting markets with deep liquidity and high trading activity . That conditional is the whole comparison. RedStone's dynamic PT oracle uses one- to four-hour windows specifically to make short-burst manipulation prohibitively expensive , and Chaos Labs has argued that feeds pulled directly from Pendle AMM data are structurally vulnerable once pools become imbalanced or sparse .

Duration is the quiet variable. A PT weeks from redemption is anchored by pull-to-par — implied yield has little room to reprice before the token converges to $1. A PT maturing Dec. 10, 2026 has months of yield to re-rate, so pushing implied APY from roughly 11% to above 20% moved the principal side about 3% . An 8.5% buffer absorbs that; a 1.007 health factor does not.

Morpho's documentation is explicit that oracle choice is made at market creation and defines the market's risk profile, and that the protocol itself is oracle-agnostic . The comparison above is therefore not a ranking of protocols but of parameter sets. Read the row that matches the market you are entering.

Who's Actually at Risk: A Fit Guide for Loopers, LPs, and Passive Lenders

Risk in this event was not distributed evenly — it landed almost entirely on one cohort. Leveraged loopers holding PT-reUSD near the 91.5% liquidation LTV absorbed the full loss, while lenders in the same markets were made whole with zero realized bad debt across both the USDC and USDT markets . That asymmetry is structural, not incidental: in a Morpho market, the borrower's collateral is the first and usually only buffer, so a 2.8% adverse move on a position with a health factor near 1.007 ends in forced repayment before any lender is touched .

Three groups face materially different exposures to the same market, and each should read this incident differently:

• Leveraged loopers (highest risk). Depositing PT, borrowing a dollar asset, buying more PT and repeating compresses the distance to liquidation with every turn. This cohort took 100% of the $36.14 million in repaid debt and 38.6 million seized principal tokens over a 14-minute window on Aug. 25, 2026 . The liquidation incentive at 91.5% LLTV works out to roughly 2.6% under Morpho's published formula, so the loss to the borrower is the seized collateral spread, not a headline penalty .
• Passive vault depositors (indirect, low-visibility risk). Depositors who route funds through curated vaults do not choose the loop, but they inherit the market's oracle and LLTV choices. A Morpho governance RFC filed before this incident had already argued for continuous collateral transparency — including leverage-loop and liquidity-deterioration alerts — warning that passive depositors can be silently exposed to loopers' leverage . Here that exposure did not convert into loss, but the gap it identified was real.
• Pendle LPs in thin, long-dated pools. No intent is required to move these markets. The Dec. 10 reUSD pool held roughly $8.97 million of liquidity against a Morpho market carrying about $67.5 million of PT collateral and $52.2 million of borrows . An LP providing depth in that pool is the counterparty to any large directional flow that reprices implied yield — which moved from roughly 11% to above 20% in nine minutes .

Before opening or holding a PT loop, run a short pre-trade check: confirm the oracle's averaging window (Pendle's own integration guidance cites 900 seconds as typical and 1,800 seconds for others, explicitly conditioned on deep, actively traded markets) ; compare the pricing pool's depth against total borrows in the lending market it feeds; and size the health-factor buffer above the largest single-trade move that pool can absorb, not the average day. Steakhouse's post-incident note put the vulnerability threshold at health factors below 1.03 — a useful floor, not a safe target.

The Disputed Question: Was This Manipulation or Just Thin Liquidity?

No protocol involved has publicly confirmed manipulation as fact. The evidence gap sits between one onchain analyst's attribution and the affected teams' more cautious framing: 0scar alleges the YT buyer and the liquidator were the same entity, citing repeatedly shared gas payments, and estimates realized gains of at least $360,000 from pairing newly acquired YT-reUSD with liquidated PT to redeem the underlying . The Defiant, which reviewed the claim, noted the profit figure was not independently confirmed . That distinction matters for anyone trying to price the risk of a repeat.

The mechanical case for intent is circumstantial but specific. The 11 Pendle trades ran between 04:28 and 04:37 UTC, and the 33 liquidation events followed between 04:37:47 and 04:51:23 UTC — a gap of under a minute between the last buy and the first seizure . Wallet 0x854e…690d, flagged by blockchain security firm PeckShield, bought in a rapid burst and then exited . Against that, the liquidation incentive itself was thin: Morpho's formula, LIF = min(1.15, 1/(β×LLTV+(1−β))) with β=0.3, produces only about a 2.6% bonus at a 91.5% liquidation LTV . Any profit thesis therefore rests mostly on the PT-redemption leg, not on the liquidation premium — which is why the $360,000 estimate remains an estimate.

The protocols' framing is narrower and, so far, uncontested on the technical facts. Pendle stated that the affected PT-reUSD/USDC market was deployed by Steakhouse on Morpho and that the oracle was configured correctly . Steakhouse described positions with health factors below 1.03 as vulnerable and put the PT asset move at 2.8% on high volume . Neither treated the cascade as a bug; both treated it as thin-liquidity, high-LTV loop risk behaving exactly as designed. Re Protocol went furthest toward the manipulation question without answering it.

"A market-price movement in the Pendle PT-reUSD oracle used by a third-party Morpho market… we are investigating whether the PT market price was intentionally manipulated and are working with the relevant teams on a safer oracle configuration," — Re Protocol, issuer of reUSD (source: CoinDesk, 2026-08).

Re Protocol also confirmed reUSD itself was unaffected — no depeg, no contract exploit . Morpho had issued no public statement at the time of reporting , consistent with its documented position that Morpho is oracle-agnostic and that oracle choice is made at market creation .

For a trader, the practical answer is that the distinction changes attribution, not exposure. Whether the $320,000 buy was an engineered attack or an opportunistic trade that happened to clear a threshold, the same parameter stack — a 15-minute TWAP over a roughly $8.97 million pool feeding a market with $52.2 million of borrows at 91.5% LLTV — produced the same $36.14 million outcome . Underwrite the configuration, not the intent.

What Changes Now: Oracle Fixes and Structural Lessons for DeFi Lending

The remediation debate after Aug. 25, 2026 has converged on three parameter-level fixes rather than code rewrites: longer time-weighted average price windows, multi-source pricing, and circuit breakers that halt liquidations when a feed deviates abnormally . RedStone's dynamic PT oracle already uses one- to four-hour averaging windows specifically to make the cost of moving a feed prohibitive relative to any liquidation gain , and Chaos Labs has argued that feeds read directly from Pendle AMM state are structurally fragile once a pool becomes imbalanced or sparse . Pendle's own integration guidance already conditions its 900-second default on selecting markets with deep liquidity and high trading activity — a condition a long-dated December 2026 pool did not meet .

Monitoring was also flagged in advance. A Morpho governance RFC filed before the incident called for continuous collateral transparency, explicitly naming leverage-loop detection and liquidity-deterioration alerts, and warned that passive vault depositors can be silently exposed to loopers' positions they never chose . Morpho's own risk documentation had likewise stated plainly that no oracle is immune to manipulation capable of producing liquidations or bad debt . The warnings existed; the parameters did not reflect them.

That is the core lesson. Nothing here broke. The oracle followed its specification, the liquidation engine cleared 33 events in under 14 minutes, and lenders were repaid in full with zero realized bad debt across both the USDC and USDT markets . What failed was composition: a 91.5% liquidation LTV calibrated for a deep market, stacked on a 15-minute averaging window calibrated for a liquid pool, applied to a thin, long-dated principal token . Each setting was defensible alone. Together they left roughly $52.2 million of borrows priced off an $8.97 million pool .

The lasting reframe for anyone holding PT as collateral: before maturity, a principal token is not a proxy for the dollar asset beneath it. It is a claim whose mark depends on the liquidity and implied-rate dynamics of the Pendle market that prices the PT/YT pair . Concrete takeaway: before opening or rolling a looped PT position, check three numbers — the oracle's averaging window, the pricing pool's depth against total borrows in that market, and your own distance to the LLTV line . If the pool is smaller than the debt it prices, treat a sub-1.05 health factor as a position already in motion.

Frequently asked questions

Did the $36 million Morpho liquidation involve a hack or exploit?

No. No smart contract was exploited, ETH did not crash, and reUSD did not depeg. The cascade came from an oracle repricing after a concentrated trade in the paired Pendle yield token pushed the principal token's market price below its scheduled accretion curve . Re Protocol described the event as "a market-price movement in the Pendle PT-reUSD oracle used by a third-party Morpho market" and confirmed reUSD itself was unaffected . Pendle stated the affected PT-reUSD/USDC market, deployed by Steakhouse on Morpho, was configured correctly . The failure was one of parameter composition, not code.

How did a $320,000 trade cause $36 million in liquidations?

Pendle splits a yield-bearing asset into a principal token (PT) and a yield token (YT), so their prices move like a seesaw — bidding YT up mechanically reprices PT down . Between 04:28 and 04:37 UTC on Aug. 25, 2026, one wallet ran 11 consecutive trades converting roughly $320,000 of SY-reUSD into more than 9.5 million YT-reUSD, lifting implied annual yield from about 11% to above 20% and dragging PT-reUSD down roughly 3% . Onchain analyst 0scar described about 5.4 million PT-equivalent flow into a pool holding roughly 3.1 million PT, dropping the 15-minute TWAP to 0.9647 . That feed priced a Morpho market carrying roughly $67.5 million of PT collateral against $52.2 million of borrows, all sitting behind a Pendle pool of about $8.97 million .

Did lenders on Morpho lose money in the PT-reUSD liquidation?

No. The Defiant's review of Morpho API data counted 33 liquidation events between 04:37:47 and 04:51:23 UTC, repaying $36.14 million of debt and seizing 38.6 million principal tokens — $35.19 million in the USDC market and $956,000 in the USDT market, with zero realized bad debt in either . Losses fell entirely on leveraged borrowers whose collateral was seized at a discount, not on lenders or passive vault depositors. Rounded totals reported elsewhere ranged from roughly $36.4 million to $36.39 million . Morpho's liquidation incentive formula, LIF = min(1.15, 1/(β×LLTV+(1−β))) with β=0.3, produces only about a 2.6% bonus at a 91.5% LLTV .

What is a Pendle principal token (PT) and why does it liquidate differently?

A Pendle principal token is a zero-coupon-bond-like claim redeemable for the underlying asset at maturity — for PT-reUSD-10DEC2026, that date is Dec. 10, 2026 . It liquidates differently from a plain dollar-denominated asset because, before maturity, its price reflects two things: the underlying asset and the liquidity and rate dynamics of the Pendle market that prices the PT/YT pair. A PT can therefore fall several percent on the oracle while the underlying holds its peg exactly, as reUSD did here . Third-party risk providers have flagged this structure directly: Chaos Labs argues feeds pulled from Pendle AMM data are structurally vulnerable when pools become imbalanced or sparse .

How can traders avoid getting liquidated in the next PT oracle event?

Check four numbers before opening or rolling a looped PT position. First, the oracle's averaging window: Pendle's integration guidance names 900 seconds as typical and 1800 seconds for other markets, explicitly conditioned on choosing deep, actively traded pools , while RedStone's dynamic PT oracle uses one- to four-hour windows to make manipulation prohibitively costly . Second, pool depth against total borrows in that market. Third, LLTV relative to time to maturity — PT-reUSD markets against USDC, USDT and AUSD all ran a 91.5% LLTV . Fourth, your own buffer: Steakhouse said positions with health factors below 1.03 were vulnerable, so treat that level as already in motion rather than as a target . Morpho's own documentation notes that oracle choice is fixed at market creation and defines the market's risk profile, and that no oracle is immune to manipulation .

Enjoyed this article? Subscribe to get new stories by email whenever they're published.

Subscribe