Block by Block: How the rsETH Exploit Rattled Aave Lending Markets

Introduction

The mid-April rsETH exploit caused the largest shock ever to the Aave protocol’s lending markets, and to DeFi as a whole. In its path it left markets around stablecoins and wrapped ether (WETH) naturally (and synthetically) frozen. As the market became aware of the exploit, users rushed to withdraw their assets, which ultimately pushed utilization rates towards 100% and prevented additional withdrawals.

The following analysis takes a granular, block-by-block look into how lending markets unwound on Aave V3 Core on Ethereum from the time of the rsETH exploit. It uses the block of the attacker’s rsETH spend approval transaction (block height 24908299 which landed on April 18 at 17:38:23 UTC) as the anchor block and looks forward toward the present. The snapshot period concludes at block height 25016299 which landed on May 3 at 18:35:23 UTC.

This report is an extension of Galaxy Research’s previous writeup covering Aave’s loan book and individual loan health and risk overviews. It draws on Galaxy Research's proprietary view of Aave markets to examine how the exploit affected the largest app in decentralized finance.

Full Market View

Before getting into the more in-depth, asset-specific data, let’s look at how the largest pools on Aave reacted to the rsETH exploit in aggregate. This analysis, and all others succeeding it, are divided into two buckets. The first observes the initial six hours from the anchor block. The second captures the full timeframe from the anchor block to present. The former is used to more closely observe the immediate reaction by the market as knowledge of the exploit proliferated, while the latter is used to observe the initial reaction plus how the dust settled.

For legibility, the report only covers assets that had more than $100 million worth of deposits at the time of the anchor block. The prices used, and all other metrics covered, come directly from public, onchain Aave contracts. We do not blend in live market feeds or offchain overlays or incorporate any information outside of how Aave manages its markets.

First Six Hours from Exploit

In the first six hours following the rsETH exploit, 1.815 billion units worth of net stablecoin supply fled Aave. USDT saw the largest outright number of native units leave the market in aggregate over this period: 807.7 million USDT (accounting for 18.27% of the USDT supplied to app). RLUSD saw the largest relative share of supply leave the market in aggregate: 83.34% (made up of 303.57 million RLUSD). GHO, the Aave-issued stablecoin, saw a 9.25% decline in units issued through the Aave V3 facilitator, leading to the burn of 10.427 million units.

WETH, the most borrowed asset on Aave by U.S. dollar value, saw a net 325,370 units leave the market. This accounted for 10.82% of the total amount of WETH supplied to Aave at the time of the anchor block.

On the bitcoin front, users withdrew: a net 6,255 cbBTC (tokens representing BTC held on Coinbase), or 26.17% of total cbBTC supplied as of the anchor block; 2,002 WBTC (wrapped BTC custodied at BitGo), 4.87% of the total supplied as of the anchor block; 351 Lombard Staked BTC (8.65% of LBTC supplied as of the anchor block); and four units of Threshold’s tBTC (0.15% of total tBTC supplied as of the anchor block). All told, a net 8,612 units of bitcoin-based assets left Aave in the first six hours (a 12.04% decline).

Of the observed cohort of assets, stablecoins saw the largest relative supply flight, losing 17.04% of their supply in the first six hours following the exploit; bitcoin-based assets saw the second largest flight, losing 12.04% of the total supplied; and WETH the saw the third highest flight, losing 10.82% of its supply. While WETH didn’t see as much supply lost over this period, the share it did lose pushed utilization to 100% in short order. The following sections will highlight why this was problematic for Aave, for depositors of WETH, and for users who had borrowed WETH for looping trades.

On the borrow side of the application, 371.851 million units worth of stablecoin borrows were closed in aggregate. This represents 5.12% of borrows closing. RLUSD saw the largest relative share of stablecoin borrows close at 80.83%. USDC saw the largest outright number of borrows close in aggregate at 158.02 million units.

Notably, WETH saw a net increase of 17,575 units worth of borrows. This is due to the exploiter borrowing 52,441 WETH against the exploited rsETH across four transactions [1][2][3][4]. Backing out the exploiter’s borrows, Aave saw a decline of 34,866 WETH borrows.

From Time of Exploit to Present

We get a full view of the exploit’s impact looking at the period between the anchor block (block height 24908299, which landed on April 18 at 17:38:23 UTC) to the last block of the snapshot (block height 25016299, which landed on May 3 at 18:35:23 UTC).

In total, 5.504 billion units of stablecoins left Aave markets. RLUSD saw the largest relative decline, with 94% of the RLUSD supplied to Aave leaving the application. USDT saw the largest outright supply flight of 2.469 billion USDT (55.85% of the total supplied).

For bitcoin-based assets, cbBTC saw the largest relative supply flight. Between the anchor and end of snapshot blocks, 47.94% of cbBTC (11,459 units) left Aave. WBTC saw the largest outright number of units withdrawn at 11,713 WBTC (28.47% of the total supplied). All –in –all, a net 25,450 units of bitcoin-based assets left Aave in the wake of the exploit (a 35.59% decline).

For WETH, 943,327 units left Aave. This accounts for nearly one third of all WETH supplied to Aave as of the anchor block.

Of the observed cohort of assets, 3.108 billion units worth of stablecoin loans (43.65% of outstanding borrows of these assets) were closed. Bucking the trend, Aave’s stablecoin, GHO, added 12.284 million units of net new borrows over the period. Additionally, 632,181 units of WETH borrows (accounting for 23.72% of the asset’s total) were closed. Notably, the Ethereum unstaking queue climbed by roughly 530,249 ETH from the day of the exploit to the peak on May 2. This suggests some portion of the recent spike in the unstaking queue is due to users unwinding their looped positions that became unprofitable once WETH borrow rates spiked on Aave.

Individual Asset Views

The following section details trends in utilization, supplies, and borrows on an asset-by-asset basis. Like the section above, it observes the first six-hour window following the exploit and the full view from exploit to present. It uses the same cohort of assets (those with supplied amounts greater than $100 million at the time of the anchor block) that are pictured in the tables above.

Stablecoins

First Six Hours from Exploit

USDT saw the greatest supply flight outright of all observed stablecoins. There were a few notable early USDT withdrawals that accounted for more than half a billion USDT within two hours of the exploit taking place at approximately 17:38:23 UTC on April 18:

• A 59,714,485 USDT withdrawal from Aave into Maple Finance’s syrupUSDT, on April 18 at 18:50:23 UTC. This transaction was the earliest large USDT withdrawal from Aave following the exploit.

• A 200 million USDT withdrawal that day at 19:12:59 UTC from an address tagged by the Etherscan block explorer as belonging to the Huobi exchange.

• A 50 million USDT withdrawal from a Huobi-tagged address at 19:14:35 UTC.

• A 285,111,407 USDT withdrawal in a transaction from the Spark Liquidity Layer contract at 19:15:47 UTC.

For USDC, the larger withdrawals took a bit longer to starting rolling in. Notably, there were the following transactions within the first three hours of the exploit:

• A 100,195,614 USD withdrawal on April 18 at 19:19:47 UTC.

• A 103,135,229 USDC withdrawal that day at 19:59:35 UTC.

• A 30,047,319 USDC withdrawal at 21:26:59 UTC.

RLUSD is the most interesting of the cohort because more than 80% of the tokens supplied to Aave net fled within the first six hours after the exploit. Over this period, there were 63 withdrawal transactions. The largest was a 113,331,405 RLUSD withdrawal from an untagged wallet. This single user accounted for 37.23% of all the RLUSD withdrawn from Aave over the first six hours after the exploit.

Stablecoin borrows only came down marginally in aggregate over the first six hours. USDC had the most net loan repays over the period. Below are some notable transactions over the period:

• A 30,000,000 USDC repay on April 18 at 19:49:11 UTC

• A 116,200,079 USDC repay that day at 20:07:23 UTC.

• A 27,060,488 USDC repay at 20:41:47 UTC.

• A 14,980,376 USDC repay at 22:52:23 UTC.

Most of the USDT repays were smaller transactions. Notably, however, there was a 26,896,666 USDT repay on April 18 at 19:24:59 UTC from an address tagged by Etherscan as belonging to Abraxas Capital Management’s Heka Funds.

The chart below tracks the utilization rates of the observed cohort of stablecoins. At the time of the exploit, stablecoin utilization rates were mostly below their target utilizations. This provided a cushion for these markets as users began withdrawing. Over the first six hours, stablecoin utilization moved by the following amounts:

• DAI: 79.98% to 82.22% (+2.24%)

• GHO: 68.29% to 61.97% (-6.32%)

• PYUSD: 54.62% to 46.89% (-7.74%)

• RLUSD: 33.69% to 38.74%, but briefly hit 100% utilization before subsiding (+5.06%)

• USDC: 76.97% to 82.72% (+5.85%)

• USDe: 57.5% to 62.08% (+4.59%)

• USDT: 77.4% to 93.27% (+15.87%)

• USDtb: 48.57% to 79.23% (+30.66%)

All in all, stablecoin utilization rates largely stayed below to slightly above each of their respective target rates over the first six hours following the exploit.

From Time of Exploit to Present

In the days following the exploit, stablecoin supplies on Aave V3 Core gradually declined before stabilizing around 4.5 billion native units. Stablecoin deposits leveled on April 24, around one week after the exploit.

The story is nearly identical for stablecoin borrows. There was a gradual grind lower for the first week following the exploit before this metric stabilized around 4 billion units.

The next chart observes stablecoin utilization rates over the observation period. While utilization rates didn’t hit 100% in the first six hours, eventually a number of them reached their max utilization. Using a 100-block moving average (MA), it took:

• USDT: 2,513 blocks to reach >99% utilization (8h 24m)

• USDC: 3,536 blocks (11h 49m)

• PYUSD: 4,005 blocks (13h 23m)

• USDe: 5,708 blocks (19h 04m)

• DAI: 12,758 blocks (42h 39m)

• RLUSD: did not reach 100-block MA utilization > 99% at any time in the observed window

For comparison, it took WETH only 678 blocks (2h 16m) to reach a 100-block MA utilization greater than 99%. Utilization rates were sticky at max utilization for a number of stablecoins:

• USDT spent 135.19 hours above 99% utilization

• USDC spent 97.82 hours above 99% utilization

• USDe spent 70.48 hours above 99% utilization

• DAI spent 15.82 hours above 99% utilization

• PYUSD spent 2.77 hours above 99% utilization

While stickiness is relative to each assets’ circumstance, WETH utilization stayed over 99% utilization using the 100-block MA for 12.7 days.

WETH

The following looks at borrows, supplies, and utilization rates for WETH. WETH was particularly important in the context of the exploit because users held rsETH collateral to loop WETH borrows on Aave.

First Six Hours from Exploit

In the first six hours after the exploit anchor block, WETH utilization rose quickly from about 88.6% to effectively 100%. The move was front-loaded: utilization climbed aggressively through hours one to three and then sat pinned near 100% for the back half of the six-hour window. The average utilization over the first three hours was about 94.0%, versus essentially 100.0% in the second three hours, which shows the market transitioned from strained to fully saturated very early and stayed that way.

The path to saturation was virtually all supply-side contraction rather than sustained borrow growth. Supplied native units fell from roughly 3.01m to 2.68m WETH (about -10.8%), with the largest drop happening in the first two hours and then a steadier grind lower. Borrowed native units initially jumped (due to the exploiter’s borrow activity) but then trended down as users unwound their positions; by the end, they were only slightly above the starting level (+0.66% net). So the headline dynamic was persistent pressure coming from liquidity leaving faster than borrow balances unwound, which kept utilization at 100% for most of the window.

From Time of Exploit to Present

Across the full analysis horizon, WETH utilization stayed structurally elevated and close to the 100% ceiling, with an average around 99.6% and only easing to about 98.47% by the end of the snapshot period. The overarching pattern is sustained tightness: even as conditions and the outlook for an exploit resolution evolved over time, utilization did not meaningfully revert to pre-stress levels, indicating liquidity remained constrained relative to borrow demand throughout most of the window.

As for balances, both sides of the trade delevered, but supply contracted significantly more than borrows. Supplied native units fell from roughly 3.01m to 2.06m WETH (about -31.4%), while borrowed native units declined from roughly 2.67m to 2.03m WETH (about -23.7%). That asymmetry leaves only a thin supply-over-borrow buffer for much of the period, which explains why utilization remains near the max despite sizable reductions in absolute borrow and supply outstanding.

WETH Borrow Cost

WETH borrow costs have been persistently above the Ethereum staking yield of approximately 2.5% with the wrapped token’s utilization rate pinned to 100%. In response to the elevated utilization rate, Aave risk curators adjusted the WETH interest rate curve on April 20 at 5:00 UTC to reduce the cost of borrowing the asset. This change is reflected in the steep drop in the borrow rate from >6% to 5%. The adjustment also included an increase in the target utilization rate to 94% to help the market find equilibrium at a higher such rate.

While the risk parameter adjustments reduced the severity of the overall impact of the market imbalance on WETH borrowers, ETH-based looping trades will remain unprofitable until utilization moves below ~95% using the current borrow curve. The chart below illustrates the current curve which includes the updates made by Aave risk curators on April 20. The vertical axis represents interest rates paid by borrowers (blue line) or the rates earned by suppliers (green) of WETH at the utilization rates on the horizontal axis. The vertical lines show the target utilization rates at present (red) and under normal conditions (dotted).

Key Collateral Assets

The following looks at the supply trends of key ETH-based and bitcoin-based collateral assets on Aave V3 Core on Ethereum. Like the above, it looks only at assets in the filtered cohort to ensure legibility.

LSTs/LRTs

First Six Hours from Exploit

In the first six hours, ETH-based aggregate collateral native units supplied (excluding WETH) declined minimally. The supply of these assets declined from about 3.80 million to 3.73 million native units (roughly -1.8%). This suggests there were few unwinds of looped ETH trades in the earliest hours following the exploit.

From Time of Exploit to Present

Over the full analysis window, the trend is a much larger cumulative de-risking. Aggregate supply fell from roughly 3.80 million to 2.63 million native units (about -31%). The withdrawal of these assets was steady over the week following the exploit, but their supplies continued to decline two weeks later. Freezes on the Aave WETH market and the logistical requirements and limitations of unwinding looping positions are likely contributors to the more gradual decline in the supplies of these assets rather than a rush for the exit.

Bitcoin

First Six Hours from Exploit

In the first six hours, bitcoin-based aggregate collateral supplied declined somewhat meaningfully. The supply of these assets declined from 71,509 to 62,897 native units (-12.04%). Taking this into account, bitcoin-based tokens supplied to Aave V3 Core on Ethereum left the platform faster than ETH-based collateral in the first six hours.

From Time of Exploit to Present

In total, 25,450 units of bitcoin-based assets left Aave (-35.59%) in the two weeks following the exploit. However, over the last week there has been an increasing (albeit small) amount of bitcoin-based assets supplied to Aave.

Conclusion

The rsETH exploit delivered an unprecedented shock to Aave markets. Within hours, a targeted attack on a single collateral asset cascaded into a broad liquidity crisis with 1.815 billion in net stablecoin supply fleeing in the first six hours, WETH utilization hitting 100% in under two and a half hours, and bitcoin-based collateral falling 12%.

The damage proved deep and lasting over the full two-week snapshot window. More than 5.5 billion units of stablecoins exited Aave, nearly a third of all WETH supply was withdrawn, and bitcoin-based collateral fell over 35%. Major stablecoin markets spent extended periods at maximum utilization (USDT for over 135 hours, WETH for more than 12 days) effectively freezing withdrawals and trapping borrowers in looped positions.

Ultimately, the exploit was a massive hit for the largest application in DeFi. However, it appears for now that the market has found a base off which it can begin to rebound. In the immediate term, unfreezing rsETH markets and working through the unwinding of unprofitable looping positions will be needed to normalize ETH markets. This can have knock-on effects on the network’s unstaking queue as users move to withdraw the staked ETH backing those trades. In the longer run, changing habits around risk curation and token construction and promoting transparency around these things will be needed to restore trust and bring liquidity back to DeFi.