Curve Finance Fee Burner Exploit: How $38K Was Stolen

8 min read
1 views
Nov 18, 2026

An attacker walked away with $38,200 from Curve Finance by tricking a fee burner with 138 fake trades. No user funds were touched, yet the way it happened should make every DeFi user pause and look closer.

Financial market analysis from 18/11/2026. Market conditions may have changed since publication.

What if a single overlooked check in a smart contract let someone walk away with tens of thousands of dollars while the rest of the protocol kept running like nothing happened? That is exactly what unfolded with Curve Finance’s fee burner in early November. An attacker managed to extract roughly $38,200 in accumulated DAO fees through a series of unauthorized trades, and the whole thing rested on a surprisingly simple signature verification gap.

How the Curve Finance Fee Burner Exploit Actually Worked

Curve Finance later confirmed that the attacker drained about $38,200 in DAO revenue on November 10. The method involved 138 fraudulent trading orders submitted against the CowSwapBurner contract on Ethereum. The money never touched user deposits, lending markets, or crvUSD holdings. It came purely from administrative fees the protocol had already collected and was preparing to convert.

I find this distinction important. In DeFi we often hear about total-value-locked disasters. This one stayed limited to protocol-owned revenue, yet it still revealed a clear design weakness that other teams should study closely.

The Signature Check That Was Never Fully Made

At the heart of the incident sat the isValidSignature function inside the CowSwapBurner. Its job was to approve orders coming through CoW Protocol. The function correctly verified that the order description met several conditions: tokens should be sold for crvUSD, proceeds should go to the FeeCollector, and the sale should return at least 500 crvUSD.

What it failed to do was confirm that the description actually matched the order being executed. The contract ignored the order hash supplied by CoW Protocol. An attacker could therefore take the signed details of a legitimate order and attach them to a completely different transaction. Once the contract had already granted CoW Protocol unlimited allowance to spend its tokens, the rest became straightforward.

Whenever the weekly trading window opened, the attacker could submit unauthorized sales and redirect the output to a personal wallet. The first fraudulent settlement hit at 00:07 UTC on November 10, just seven minutes after the window unlocked. That initial trade sold the burner’s YB token balance for USDC and sent the proceeds straight to the attacker.

The same legitimate order description was then reused for another 137 transactions covering other tokens. By 00:24 UTC, 133 of those orders had already settled. The remaining five finished over the following hours, with the last one completing at 11:10 UTC.

Why User Funds Stayed Completely Safe

Curve’s post-incident review made one point crystal clear: the attacker targeted only accumulated administrative fee balances. The protocol’s normal process moves fees from pools and markets into a burner. Those tokens are then converted into crvUSD and later distributed to the DAO.

Under the existing schedule, fee tokens arrive on Mondays, get sold during a 24-hour window that starts on Tuesday, and are distributed on Wednesdays. Even while the exploit ran, 36 legitimate orders involving 35 different tokens still produced 97,317.74 crvUSD for the FeeCollector. The attacker’s 138 forged orders, covering 135 tokens, returned 38,205.76 USDC instead.

Most of the stolen balances had sat unsold for a long time because they fell below the 500 crvUSD minimum sale threshold. The median fraudulent order was worth only about $254. Just two of them exceeded $500. For tokens that had been traded before, the median time since their last legitimate sale stretched to 161 days.

Curve examined every one of the 6,558 orders settled through the burner since its deployment in June 2024. Only those 138 unauthorized transactions sent proceeds somewhere other than the FeeCollector. That level of historical review gives reasonable confidence that the vulnerability had not been exploited earlier.

The Path the Stolen Funds Took

Within four hours of the first trade, the attacker swapped 37,611.76 USDC into ETH, spread the funds across four addresses, and bridged them to Arbitrum. Roughly 37,229.89 USDC later reached Hyperliquid, where it was exchanged for 65.41 units of a Monero-linked asset. As of mid-November that balance remained untouched.

A smaller leftover of 594 USDC was converted to ETH on November 15, and 0.3 ETH was deposited into Tornado Cash. The mix of a privacy-focused bridge asset and a classic mixer suggests the attacker wanted multiple layers of obfuscation, even if the absolute amounts were modest by current DeFi standards.


How Curve Responded in Real Time

An external vulnerability report arrived on November 10. The Emergency DAO reacted quickly: fee collection and trading on Ethereum were disabled at 13:59 UTC the same day. Similar pauses followed on Arbitrum and Base at 16:55 UTC. Ethereum fee conversion remains offline while balances continue to accumulate.

On November 13 a governance proposal went live. It aims to replace the vulnerable burner with a descending-price auction system that properly verifies the actual order hash. Voting is scheduled to close on November 20. In the meantime, on November 17 the Emergency DAO moved residual token balances from the old burner contracts on Ethereum, Base, and Arbitrum back to their respective FeeCollector contracts.

From an operational standpoint the containment looks solid. The team limited further exposure, protected the remaining fee inventory, and opened a clear path toward a more robust replacement mechanism.

What This Incident Reveals About DeFi Design Assumptions

Perhaps the most interesting aspect is how narrow the window of opportunity really was. The attacker needed the trading window to be open, needed leftover balances that had not yet been sold, and needed the signature check to remain incomplete. All three conditions aligned for a short period.

In my view this episode underscores a recurring lesson: allowances and signature validation are high-stakes surfaces. Granting unlimited spending rights to an external protocol is convenient, yet it amplifies the cost of any downstream verification error. Teams that rely on order-description checks without binding them to the concrete order hash leave an opening that sophisticated actors will eventually test.

It is also worth noting that Curve has dealt with other security events in the past, including infrastructure issues and market-specific incidents. None of those earlier events involved this particular fee-conversion logic. The fact that the burner had processed thousands of legitimate orders over many months without prior abuse suggests the flaw was not trivial to discover or exploit until someone specifically looked for it.

A Closer Look at the Numbers

The raw figures help put the event in perspective. Here is a simplified comparison of legitimate versus unauthorized activity during the incident window:

CategoryOrdersTokens InvolvedValue
Legitimate sales363597,317.74 crvUSD
Unauthorized sales13813538,205.76 USDC
Median unauthorized order——≈ $254

Most of the stolen value came from small leftover balances that had been sitting idle. That detail is both reassuring and slightly unsettling. It shows the protocol’s minimum-sale threshold had kept many tokens from being processed, yet those same idle balances became convenient targets once the verification gap was found.

Lessons That Extend Beyond This Single Contract

Anyone building or auditing fee-management systems should treat order-hash binding as non-negotiable. Description checks alone are insufficient when an external settlement layer can present arbitrary orders. Unlimited allowances should also be reviewed periodically; time-limited or amount-limited approvals reduce the blast radius of similar bugs.

Monitoring for anomalous settlement destinations is another practical defense. Curve’s later review of thousands of historical orders made the anomaly obvious in hindsight. Real-time alerts on proceeds leaving the expected FeeCollector address could have shortened the window of exposure.

Finally, the decision to move residual balances out of the old burners after the incident is a clean operational practice. Leaving funds in a known-vulnerable contract, even after trading is paused, creates unnecessary residual risk.

Where Things Stand Now

As of the latest available information, the stolen funds have not been recovered. The largest portion sits in a Monero-linked asset on Hyperliquid, while a small ETH amount went through a mixer. Ethereum fee conversion remains paused, and the governance proposal for a new auction-based burner is still open for voting until November 20.

The overall impact on Curve’s core operations appears limited. User deposits were never at risk, and legitimate fee conversion continued for the bulk of the available tokens. Still, the episode serves as a concrete reminder that even mature protocols can harbor subtle verification gaps in administrative contracts.

I’ve followed enough of these incidents to know that the interesting part is rarely the dollar amount. It is the precise sequence of assumptions that turned out to be incomplete. In this case the assumption that an order description would always match the executed order proved costly. Fixing that single missing check should close the door for good.

Why Small Balances Became Attractive Targets

One detail that stays with me is how many of the drained positions were essentially dust that had accumulated over months. The median time since last sale was more than five months. Those balances were too small to meet the 500 crvUSD threshold under normal operation, so they simply waited. Once the signature gap appeared, they became easy pickings.

This pattern appears in other protocols as well. Idle residual balances often receive less operational attention than active large positions. Yet from an attacker’s perspective they represent low-visibility, high-certainty targets when a verification flaw exists. Periodic sweeps or lower thresholds for residual sales might reduce the surface area available to similar attacks in the future.

The Role of External Reporting

Credit belongs to the external party who reported the vulnerability on November 10. Without that report the attacker might have continued draining additional residual balances in subsequent weekly windows. The Emergency DAO’s rapid response—disabling the burner within hours—limited further damage. Coordination between external researchers and internal emergency processes remains one of the stronger defensive layers in DeFi.

It is also notable that the team chose transparency. Publishing a clear timeline, the exact number of unauthorized orders, the destination of funds, and the planned remediation helps the broader ecosystem learn. Quiet patches leave other projects guessing; detailed disclosures accelerate collective improvement.

Looking Ahead to the Replacement Mechanism

The proposed descending-price auction system is designed to verify the actual order hash rather than relying solely on a description. That change directly addresses the root cause. Auctions also introduce price discovery that may improve conversion efficiency compared with the previous fixed-window approach, though the primary goal remains security.

Whether the community approves the proposal on schedule will determine how quickly the new system can be deployed. Until then, fees continue to accumulate on Ethereum while conversion stays paused. The residual balances have already been moved to safer contracts, so the immediate risk of further loss appears low.

In the longer term, incidents like this tend to raise the baseline for what is considered acceptable verification hygiene. Other protocols that use similar order-settlement patterns will almost certainly review their own signature logic after reading the details. That secondary effect may ultimately prove more valuable than the $38,200 itself.

Final Thoughts on Containment and Transparency

The Curve Finance fee burner exploit was neither the largest nor the most complex attack of the year. It was, however, a clean illustration of how a single incomplete check can turn administrative convenience into a loss vector. The attacker needed no sophisticated flash-loan machinery or novel economic attack. A mismatched order description and an open trading window were enough.

What stands out is the speed and clarity of the response. Trading was halted, residual funds were secured, and a governance path toward a safer design was opened within days. Those steps matter more than the absolute size of the loss when evaluating operational maturity.

For anyone watching DeFi security, the takeaway is straightforward: never assume that a description check equals an order-hash check. Bind them explicitly. Limit allowances when possible. Monitor settlement destinations in real time. And when something does go wrong, move residual balances out of the compromised contracts without delay.

The $38,200 is gone for now. The design lesson remains available to every team that cares to study it.

❝
Opportunities don't happen, you create them.
— Chris Grosser
Author

Steven Soarez passionately shares his financial expertise to help everyone better understand and master investing. Contact us for collaboration opportunities or sponsored article inquiries.

Related Articles

?>