Nine Months in the Making: Venus Protocol's Fourth Supply-Cap Exploit Leaves $2.15M in Bad Debt

On March 15, 2026, an attacker who had spent roughly nine months quietly building an 84% share of Venus Protocol's supply cap for Thena's THE token triggered a Mango Markets-style price manipulation on BNB Chain. The method — a so-called donation attack — sidestepped the supply cap entirely, fed a recursive borrowing loop against shallow liquidity, and pulled out $3.7 million in borrowed assets before the position unwound into $2.15 million of bad debt.

It is the fourth major security incident Venus has faced since 2021, and the second time the underlying flaw has been collateral-inflation via donation, a pattern the protocol had already lived through once before on its own ZKSync deployment twelve months earlier, where it absorbed $717,000 in losses.

What makes this instance notable is that the exact mechanism — donations bypassing supply-cap enforcement — was documented in Venus's own Code4rena audit contest in 2023. The team's response at the time was to classify it as "supported behavior with no negative side effects" and leave it unaddressed.

Independent researcher William Li identified the attack while it was still unfolding on March 15, publicly named the attacker's wallet before Venus said anything, and closed a short position for a $15,000 profit as the exploit collapsed.

Despite pulling $5.07 million in assets out of the protocol, the attacker's on-chain accounting suggests they may have ended up with little or nothing to show for it, once accumulation costs are factored in. Venus, meanwhile, is left owing governance an explanation for a $2.15 million hole in its books.

01How the response unfolded

Li's public warning came first: "Hi Venus Protocol, I think there is some suspicious activity going on with your vTHE pool. Someone is 'donating' to jail-break the THE supply cap and performing a Mango-market like attack. Please pause your pool ASAP!"

This wasn't a lucky guess. Li had previously modeled this category of attack in a 2023 academic paper examining role-play attack strategies in DeFi, and recognized the pattern the moment THE's price began moving abnormally — posting the attacker's address before Venus Protocol had issued any statement.

Attacker address: 0x1A35bD28EFD46CfC46c2136f878777D69ae16231

Blockaid's community alert described it as "delegated borrowing abuse." Roughly two hours after Li's initial post, Venus issued its first statement: "We have identified unusual activity involving the $THE pool and are actively investigating. At this time, only the $THE and $CAKE markets appear to be affected."

An hour after that, borrows and withdrawals on THE were paused. Two hours later, Venus zeroed out the Collateral Factor across six additional markets — BCH, LTC, UNI, AAVE, FIL, and TWT — each flagged because a single wallet controlled more than 60% of the supplied collateral. lisUSD was added shortly after under the same criteria.

Thena moved to distance itself from the incident: "THENA Smart Contracts are safe. A malicious actor targeted the THE market on Venus Protocol. Your funds and positions on THENA remain SAFU."

By the next morning, Venus published its account of events: the attacker had spent nine months accumulating THE to build an outsized supply position, bypassed the cap through direct contract transfers, ran a recursive borrowing loop, and generated bad debt. The statement maintained that oracles did not fail and that Venus Flux was unaffected.

That oracle claim needs context. The official post-mortem, published March 17, confirms that Venus's BoundValidator actually rejected the spiking price feed for roughly 37 minutes before both price feeds converged and the manipulated rate was accepted. Technically, the oracle didn't fail — it held out for a while before being overwhelmed.

Venus's statement included one notable admission buried among the damage control: "This is a gap in our code we are working to close." By that point, Li had already closed his short for the $15,000 profit noted above.

02Nine months of quiet accumulation

Most DeFi exploits are assembled in hours or days. This one took nine months to build.

According to Venus's post-mortem, a wallet began receiving ETH in June 2025 — ultimately 7,447 ETH, worth roughly $16.29 million, arriving across 77 separate Tornado Cash transactions. That ETH was deposited as collateral on Aave, used to borrow approximately $9.92 million in stablecoins, and those funds were routed through intermediary addresses to buy THE on the open market over time.

No rule was broken during this phase. There were no alerts. The wallet simply accumulated, eventually representing 84% of Venus's 14.5 million THE supply cap.

Primary attacker address: 0x1a35bd28efd46cfc46c2136f878777d69ae16231

Notably, community members had already flagged this address as suspicious before March 15, but Venus reportedly took no action, since no individual transaction had technically violated any rule.

At some point the deposit phase ended and direct token transfers began. A second wallet was also involved in the operation, identified by Quill Audits, though its precise function — helper wallet, liquidation front-runner, or beneficiary — isn't fully clarified in public analyses.

Second attacker address: 0x43c743e316f40d4511762eedf6f6d484f67b2f82

03The mechanism: a donation attack

Venus enforces its supply cap in a single location: the mint() function, which is the standard deposit path. Any deposit that would breach the cap causes the transaction to revert.

What mint() does not police is a raw ERC-20 transfer() sent directly to the vTHE contract — a path that circumvents mint() altogether.

Tokens moved this way still arrive in the contract and are still counted by balanceOf(). Because the vToken exchange rate is derived from that balance, it rises accordingly — but the supply cap check, which only fires on mint(), never sees the deposit.

The attacker's own vTHE token balance never changed and no new vTokens were issued. Instead, the same fixed number of tokens went from representing about $3.3 million in collateral to more than $12 million simply because the underlying exchange rate had been inflated 3.81x by tokens the protocol wasn't tracking against the cap.

This is the donation attack in essence: send tokens directly to the contract, inflate the exchange rate, and borrow against collateral the cap was specifically designed to prevent. It's a known weakness across Compound V2-style forks, and it had already been flagged in Venus's 2023 Code4rena audit — where it was dismissed as "supported behavior with no negative side effects."

How the cap was overrun

The breach progressed in three stages:

• 12.2M THE supplied through normal deposits — 84% of the cap, still within limits
• 49.5M THE — 341% of the cap — added via direct donation transfers
• 53.2M THE — 367% of the cap — at peak, just before liquidation

Six wallets coordinated the donation transfers, sending a combined ~36 million THE directly into the vTHE contract in the opening transaction.

Attack contract: 0x737bc98f1d34e19539c074b8ad1169d5d45da619

The attack contract itself held no collateral; it operated instead with delegated permission to draw against the borrowing power of the 0x1a35 wallet. Splitting the roles this way let the attacker run the extraction from a separate address than the one holding the collateral base.

With a collateral position 3.67 times the supposed maximum, the attacker now had borrowing power far beyond what the thin on-chain liquidity for THE could actually support.

04Running the loop

The mechanics were straightforward and repeatable: deposit THE as collateral, borrow CAKE, BNB, BTCB, and USDC against it, then use those borrowed assets to buy more THE on the open market — a market with thin liquidity, so each purchase moved the price sharply. The newly bought THE was then transferred directly into vTHE, again bypassing the cap, followed by a wait for the TWAP oracle to catch up before repeating the cycle.

THE's spot price was driven from roughly $0.26 to nearly $4 on the thinnest available DEX liquidity.

Venus's Resilient Oracle setup, which cross-references multiple price sources including RedStone and Binance, didn't simply track the manipulated price upward. Its BoundValidator component rejected the spiking Binance feed for approximately 37 minutes while the divergence between feeds remained too wide.

Only once the attacker sustained enough buying pressure across multiple venues to pull both feeds into alignment did the BoundValidator accept the new rate — settling at roughly $0.51, far below the $4 spot peak, but still nearly double the pre-attack price and enough to unlock another round of borrowing.

Primary attack transaction: 0x4f477e941c12bbf32a58dc12db7bb0cb4d31d41ff25b2457e6af3c15d7f5663f

Second attack transaction: 0xce6e3eb2a28ced1ef1c2212f36736e89f647365b0dfad6c9addc4c8b31f5fb0e

At its peak, the position held 53.2 million THE as collateral against borrowed 6.67 million CAKE, 2,801 BNB, 1.97 thousand WBNB, 1.58 million USDC, and 20 BTCB. Total borrowing at peak reached approximately $14.9 million — the more widely cited $3.7 million figure reflects net assets extracted after the position was liquidated.

Hacken's analysis counted roughly 50 exploit transactions in total. Separately, the attack contract had also pre-supplied 1.58 million USDC as collateral and borrowed 4.63 million THE at 11:55 UTC during the same opening window. That secondary position began liquidating at 12:04 UTC, roughly forty minutes ahead of the main position's peak.

05The collapse

Rather than stopping after the initial extraction, the attacker kept buying THE with borrowed funds, apparently trying to force another price leg and another round of borrowing capacity. Sell pressure built, the price stalled, and the account's health factor drifted toward the 1.0 liquidation threshold.

When liquidation triggered, 53.2 million THE hit an order book with virtually no depth, and the price collapsed to $0.22 — below where it had started. The roughly $30 million in nominal collateral value that had existed on paper vanished.

Venus was left holding $2.15 million in unrecoverable bad debt: 1.18 million CAKE and 1.84 million THE that liquidation proceeds could not cover. As of publication, the extracted funds remained sitting in the attacker's wallet, with no mixer activity or apparent exit path detected on-chain.

06A warning that was documented, then dismissed

Venus's exposure to this class of exploit wasn't a blind spot — it was a known issue that had been evaluated and set aside. Code4rena's audit contest in May 2023 documented the donation attack in detail, including a working proof of concept filed as finding M-10: mint vTokens, donate underlying tokens directly to the contract, watch the exchange rate inflate, then borrow against collateral you shouldn't have had access to. The mechanics and math were both laid out clearly in the report.

The official March 17 post-mortem quotes Allez Labs directly acknowledging that the vector "was identified in a prior Code4rena audit but was assessed as having 'no negative side effects' and was not remediated." This isn't external speculation — it's Venus's own risk-management partner confirming, in the formal post-mortem, that the dismissal happened and naming it as a root cause of the incident.

The gap wasn't purely theoretical even before March 2026. In February 2025, a donation attack hit Venus's ZKSync deployment using nearly identical mechanics. The wUSDM exchange rate was manipulated through a direct ERC-4626 donation; Venus froze the affected market, and the protocol absorbed $902,159 in bad debt, later partially offset by $163,757 in liquidation fees, for a net loss of $716,789. A post-mortem was published at the time, noting that the BNB Chain Core Pool shared the same underlying vulnerability — which was not subsequently patched.

The sequence, laid end to end: auditors flag the vector in 2023 and the team dismisses it; an attacker exploits the same vector in 2025 and the team absorbs the loss; an attacker exploits it again in 2026, on the chain that was never patched, and the team describes it as "a gap in our code we are working to close."

Hacken's post-incident analysis put it directly: "Attention Compound V2 forks: Verify whether direct token transfers to your cToken contracts bypass supply-cap logic. If so, the same attack pattern may still be exploitable." Venus is itself a Compound fork, and that warning had already applied to it twice before this incident.

The community's due diligence had also flagged the attacker's accumulation wallet before March 15 — concentrated ownership, a dominant position in a low-liquidity market, and funding traceable to Tornado Cash. No action followed, because each individual deposit was technically compliant with existing rules; the rules themselves simply had a gap.

Venus's post-incident criteria for pausing markets — market cap under $2 billion, 24-hour trading volume under $100 million, DEX TVL under $40 million, and single-wallet concentration above 60% — describe the THE market's pre-attack profile precisely. Those thresholds now exist as active triggers; they did not exist as enforcement mechanisms before March 15.

Between the audit finding, the prior exploit, and the community flag, Venus had three independent signals pointing at the same underlying gap before this incident occurred.

07Did the attacker actually profit?

Venus's $2.15 million in bad debt is clear. Whether the attacker came out ahead is less so.

The on-chain figures are not straightforwardly favorable to the attacker: they entered the operation with 7,447 ETH funneled through Tornado Cash and borrowed an additional $9.92 million via Aave to finance the accumulation phase. Against that outlay, they extracted roughly $5.07 million in assets from Venus — 2,172 BNB, 1.516 million CAKE, and 20 BTCB. The position had briefly carried about $30 million in nominal collateral value on paper, all of which liquidation reduced to nothing.

As EmberCN summarized it: "He borrowed 9.92 million U (stablecoins) to stir things up, but the assets borrowed from Venus were only worth $5.07 million. Onchain alone, it doesn't look profitable." William Li told The Block much the same: "From onchain analysis, he almost didn't profit."

The official post-mortem goes a step further: Allez Labs confirmed on-chain analysis shows no net gain, then added that "the strategy could include independent wallets or CEX accounts, to long/short THE. We are coordinating with exchanges and forensic partners to pursue this line of investigation." That off-chain hypothesis is now the active line of inquiry Venus and its partners are pursuing.

As of publication, the extracted funds remain parked in the attacker's wallet, with no Tornado Cash activity, bridge movement, or other visible exit path.

There is one scenario in which the numbers reverse: if the attacker also held large short positions on THE perpetuals at a centralized exchange, shorting into the manufactured pump and collecting on the subsequent collapse to around $0.21, the on-chain loss becomes irrelevant to their overall return. Under that reading, the protocol-level damage would have been the mechanism for generating a payday elsewhere, not the payday itself — THE's price round-tripped from $0.26 to nearly $4 and back below its starting point, and anyone short from the top would have profited substantially, though such a position isn't verifiable on-chain.

Li demonstrated the smaller-scale version of exactly this trade himself: he shorted THE via a perpetual futures contract as liquidation began, closed it near $0.24, and banked the $15,000 profit noted earlier — the same playbook, disclosed openly, at far smaller scale. EmberCN raised the same possibility for the attacker: "I suspect he dominated the THE downturn through on-chain liquidations to profit from his positions on the CEX."

Nine months of preparation, 7,447 ETH seeded through a mixer, and a near-identical exploit already on record from 2025 — and the attacker may have broken even, lost money, or made their real profit through an untraceable derivatives position. Venus absorbs the bad debt regardless of which version turns out to be true.

08A pattern going back five years

This is the fourth security incident in five years for Venus, and each has followed a variation of the same underlying failure mode: collateral valuation that can be manipulated to exceed what the protocol's own risk limits were meant to allow.

• 2021: $95 million in bad debt from an XVS price manipulation.
• May 2022: $14.2 million lost when attackers exploited a stalled Chainlink oracle to borrow against artificially inflated LUNA collateral during the LUNA crash.
• February 2025: $717,000 from a donation attack on the ZKSync deployment — a structurally related exploit on a different chain that went unheeded as a warning.
• March 2026: $2.15 million, following a nine-month setup that the community had flagged, an auditor had documented, and a prior exploit had already proven viable.

Venus described the latest incident as a code gap. Li had already called it a Mango Markets-style attack before Venus said a word publicly. Hacken framed it as a warning to every Compound V2 fork running the same logic. Code4rena had flagged the mechanism back in 2023, only for it to be assessed as having no negative side effects and left unfixed.

The remediation commitments follow a familiar template — supply cap hardening, tighter collateral eligibility, additional price-monitoring safeguards — the same assurances that have followed each previous incident. This time, the list is somewhat longer: a formal governance proposal for bad debt resolution, coordination with law enforcement on the Tornado Cash funding trail, and a full re-audit of the core pool. Whether a longer list translates into a longer institutional memory remains to be seen.

ChainCatcher described Venus as the "most experienced" project in the space at handling attacks — one way to interpret five recurring incidents. Another is that five years of survival is not the same thing as five years of learning.

The post-mortem confirms the bad debt will be covered by the protocol, and outlines short-term remediation steps: reducing collateral factors for illiquid assets, introducing more conservative liquidity-weighted calibrations, and reviewing all market configurations and collateral parameters, with potential delistings under consideration.

Venus will almost certainly survive this incident, as it has the previous three. The more open question is how many other Compound V2 forks, reading Hacken's warning today, will conclude it doesn't apply to them — and how many of those will be publishing a similar post-mortem within the next six months.