More Markets WFLOW Exploit Drains $9.3M On Flow EVM

Fifteen and a half million wrapped tokens do not vanish because someone fat-fingered a transfer. They vanish because a lending market treated two similar looking assets as if they were the same kind of collateral, and an attacker understood that difference better than the code did. That is the uncomfortable starting point for the More Markets incident on Flow EVM: a reported drain of about 15.5 million WFLOW, with a detected impact near $9.3 million, and a trail of follow-up transfers that investigators were still mapping when the first alerts went out.

What Actually Happened Inside The Lending Reserve

I have covered enough DeFi incidents to know the first number is rarely the last number. Security researchers flagged More Markets, a noncustodial lending protocol built by More Labs and deployed on Flow EVM, after an attacker emptied the mFlowWFLOW reserve. The public assessment put the outflow at 15.5 million WFLOW and described the dollar figure as a detector impact, not a closed forensic total. That distinction matters. Detected impact is the smoke. Final loss is the fire once every hop, wrap, swap, and bridge is counted.

The protocol itself is not some obscure experiment sitting in a testnet corner. It follows an Aave V3 style design. Users supply assets, earn interest, borrow against collateral at variable rates, and liquidate positions that fall under the required threshold. Public materials list multiple supported markets. WFLOW sits in that set as the wrapped native asset. An Ankr liquid staking token sits there too. Those two facts, placed next to the protocol’s E Mode settings, are the hinge of the story.

Researchers said the attacker used an Ankr bonded liquid staking token together with E Mode to drain the WFLOW reserve. They published an exploit transaction, a contract deployment transaction, and a cluster of later transfers used to move funds after the reserve was emptied. They did not, at least in that first wave, claim that Ankr itself had been hacked or that the Flow chain had been taken over. The target named in the disclosure was More Markets.

The first disclosure is almost never a complete autopsy. It is a map with a few streets labeled and a lot of alleys still dark.

Why The Dollar Figure Still Feels Unfinished

People love a clean headline. Nine point three million sounds clean. Markets do not work that way after an exploit. Price at the moment of detection, slippage during exit, leftover inventory inside the attacker wallet, and assets still sitting in intermediate contracts all change the picture. If WFLOW’s local liquidity is thin, dumping 15.5 million units can crush the printed value before the last transfer confirms. If the attacker parks tokens instead of selling, the paper impact and the realized cash can drift apart for days.

That is why I keep saying the investigation is the story, not just the drain. A cluster of post-exploit transfers was identified. Those transfers are the part readers should watch. They tell you whether the attacker is sloppy, professional, or still improvising. They also tell protocol teams where to aim freezes, warnings, and exchange alerts, assuming any centralized venue is willing to listen.

A Quick Look At The Market The Attacker Walked Into

More Markets lists WFLOW with a loan-to-value ratio around 81.5% and a liquidation threshold around 83%. The Ankr FLOW staking token is listed with a loan-to-value near 78.5% and a liquidation threshold near 81%. Those numbers look conservative on a slide. In production they become a machine that assumes prices, exchange rates, and asset identities stay honest under stress.

E Mode is the amplifier. In Aave-style systems, Efficiency Mode lets highly correlated assets borrow against each other at more aggressive parameters. The idea is simple. If two tokens are supposed to move together, the protocol can safely offer more leverage. The risk is equally simple. If that correlation is an assumption rather than a hard constraint, E Mode becomes a loaded spring.

Ankr’s FLOW staking token is described in product docs as a reward-bearing liquid staking token. Users stake FLOW, receive the LST, and the token’s value relative to FLOW rises as rewards accrue while the holder’s token count stays the same. There is a ratio feed. There are separate contracts for the token, the staking pool, configuration, and that ratio certificate. On paper, that architecture is tidy. Inside a lending market, the tidy part can hide a pricing or accounting mismatch.

The Difference Between A Protocol Bug And A Chain Breach

This point will get lost in social feeds, so I want it in plain language. An application exploit on Flow EVM is not automatically a Flow network compromise. Flow EVM is an Ethereum-compatible environment sitting on Flow. Lending contracts run there because developers want EVM tooling. That design choice is convenient. It is also a reminder that EVM risk travels with the virtual machine, even when the settlement layer is a different chain.

Flow has promoted both More Markets and Ankr inside its DeFi lineup. Community reward programs even pointed users toward lending activity and liquid staking. That marketing loop is normal. Ecosystems need usage. The downside is psychological. When a featured app gets hit, casual users blur the line between app failure and chain failure. I think that blur is dangerous. It either over-punishes a network or under-punishes a protocol, and neither outcome helps depositors.

The distinction is sharper because Flow already lived through a separate security crisis in late 2025. That earlier incident hit the Cadence execution layer, not this lending market. An attacker found a way to duplicate fungible tokens and extracted roughly $3.9 million in value. More than a billion counterfeit FLOW tokens moved toward centralized venues. Validators halted the chain. A full rollback was floated, then dropped after bridge operators and other participants pushed back. The network shifted to an isolated recovery that tried to destroy counterfeit supply without rewriting every honest transaction.

Later reporting around that recovery said hundreds of millions of FLOW were returned by major exchanges and destroyed, while the network isolated most of the remaining fake supply. Accounts tied to suspicious activity faced temporary limits. Developers worked to restore Cadence and EVM functionality. Months afterward, the fallout even reached exchange-listing fights in South Korea. That history is not this exploit. It is the shadow this exploit walks into. Users remember the last fire when they smell new smoke.

How Liquid Staking Tokens Create Soft Edges In Lending Markets

Here is the part I wish more depositors would sit with. A wrapped native token and a reward-bearing staking token are cousins, not twins. WFLOW is meant to stand in for FLOW inside EVM contracts. The staking token represents staked FLOW plus a changing claim on rewards. Its exchange rate is not a static one-to-one sticker. It lives through a feed, a certificate, or an on-chain ratio. Lending protocols that treat those instruments as near substitutes in E Mode are making a bet on that feed, on the implementation that reads it, and on the economic path an attacker can take between mint, collateralize, borrow, and exit.

I have found that teams often audit the staking contracts in isolation and audit the lending fork in isolation. The dangerous room is the hallway between them. Pricing assumptions. Decimal handling. Share versus asset accounting. Whether a bonded token can be used in a way the risk parameter spreadsheet never modeled. Whether E Mode’s correlation list is a living risk control or a leftover from a launch checklist.

None of that requires me to pretend I have the internal trace of this specific attack. The first disclosure did not publish a full technical sequence. It remains unclear, from that early picture, whether the flaw sat mainly in More Markets’ implementation, in the way the Ankr asset was wired into the market, in pricing assumptions, or in the interaction of those pieces. Honesty beats a fake diagram. What we can say is the attacker needed both the bonded LST and E Mode to make the drain work, at least according to the researchers who first flagged it.

• WFLOW acted as the reserve that actually emptied.
• The bonded staking token appears to have been the wedge.
• E Mode appears to have been the leverage that made the wedge wide enough.
• Post-exploit transfers show the attacker did not stop at the reserve.

What E Mode Is Supposed To Do, And Where It Goes Sideways

If you have never clicked into an Aave-style interface, E Mode looks like a gift. Higher loan-to-value. Better capital efficiency. Less “wasted” collateral if you are looping correlated assets. For professional farmers that gift is the whole product. For a protocol, it is a concentrated risk bucket. You are telling the system that these assets will not diverge enough to leave bad debt behind.

That promise breaks in a few familiar ways. An oracle can lag. A staking ratio can be read incorrectly. A token that is “bonded” can have transfer, unwrap, or valuation properties that vanilla collateral does not. A newly deployed helper contract can mint or manipulate an intermediate position faster than liquidators can react. Sometimes the issue is not even a classic oracle hack. It is an accounting mismatch that lets one side of a pair be counted as more valuable than the reserve standing opposite it.

Perhaps the most interesting aspect is how ordinary the setup looks before it fails. Supported asset. Documented LTV. Documented liquidation threshold. Audited staking contracts somewhere in the ancestry of the token. A popular architecture copied from a battle-tested Ethereum protocol. Users see that checklist and feel safe. Attackers see the same checklist and look for the one assumption nobody stress-tested against a hostile sequence.

The On-Chain Breadcrumbs Researchers Already Flagged

Three artifacts were called out early: the exploit transaction, the contract deployment transaction, and the transfer cluster that followed. That trio is a familiar pattern. Deploy a purpose-built contract. Execute the drain. Scatter or convert the proceeds. The scatter step is where many attackers get sloppy and where responders still have a chance. If funds sit in a transparent EVM environment, labeling services can tag the cluster quickly. If they move into mixers, obscure bridges, or low-liquidity pairs, the window shrinks.

I would not assume a single destination. Attackers often split. Some tokens stay in the original wallet as a taunt. Some go to fresh addresses. Some hit a DEX in pieces so the price chart looks like ordinary selling until you zoom in. Until a final accounting lands, any claim that “the hacker got away with exactly $9.3 million in spendable cash” is premature. The reserve is empty. The cash-out is a separate movie.

Working timeline, still incomplete:
  1. Helper or attack contract is deployed
  2. Bonded LST plus E Mode interaction hits the market
  3. mFlowWFLOW reserve loses about 15.5M WFLOW
  4. Transfer cluster moves value after the drain
  5. Forensics tries to pin final wallets and realized value

What Depositors Should Do In The Next 48 Hours

If you have funds in the affected market, stop thinking like a yield tourist and start thinking like an operations person. Check whether the interface still lets you withdraw unaffected assets. Check whether a pause has been announced. Check whether your position is now undercollateralized because the reserve that backed borrowing vanished. Do not approve new spender contracts while the dust is in the air. Fresh “recovery” sites appear after every exploit. They are often the second heist.

If you are not in the market but hold FLOW, WFLOW, or the related staking token, watch liquidity, not slogans. A drained reserve can force distressed selling. Distressed selling can print a local price that has little to do with the chain’s long-term design. That local price still hurts anyone who needs to exit today.

1. Confirm official protocol channels before clicking any “reclaim” page.
2. Review allowances on the lending contracts you actually used.
3. Separate emergency exits from panic exits; slippage can be the second loss.
4. Watch whether governance or admins pause markets or update oracles.
5. Keep records of positions in case a reimbursement process appears later.

Why “Audited” Keeps Failing As A Comfort Word

Ankr has said its Flow liquid staking contracts on Cadence and EVM went through external review. That statement can be true and still leave a lending integration exposed. Audits are snapshots. They cover a scope. They do not babysit every future parameter change, every E Mode listing, every oracle route, every wrapper. I get irritated when marketing treats an audit logo like a force field. Users deserve better language. Audited means someone looked. It does not mean the economic path between two listed assets is safe under adversarial composition.

In my experience, the reports that actually help after an incident are the ones that admit composition risk. Token A is fine. Protocol B is mostly fine. A plus B plus E Mode plus a custom helper contract is a new machine. That new machine needed its own threat model. If it did not get one, the market paid for the missing page.

Flow EVM As A Convenience Layer With Imported Scars

Ethereum compatibility is catnip for developers. You keep Solidity habits, familiar tooling, and a huge hiring pool. You also import a decade of EVM exploit patterns. Flash-loan style sequencing, approval races, crafted callbacks, and parameter gymnastics travel well. A chain can have a different consensus design and still host the same class of application bugs once it offers an EVM room.

That is not an argument against Flow EVM. It is an argument for treating EVM apps on Flow with the same paranoia people finally learned on Ethereum. Liquidity mining does not cancel that paranoia. Ecosystem reward points do not cancel it either. If a protocol offers high LTV on a staking derivative and a wrapped native asset in the same efficiency basket, users should ask who is responsible when the basket tears.

The Earlier Cadence Crisis Still Shapes The Mood

Let me be careful here, because mixing two incidents is how rumors get legs. The December 2025 Cadence event was a runtime problem. A protected asset that should not have been copyable was disguised as a standard data structure and duplicated. Dozens of malicious contracts were part of a coordinated sequence. The chain halted. A global rollback was proposed, then abandoned because reversing confirmed history can duplicate balances for people who bridged during the window and punish users who did nothing wrong.

The isolated recovery that followed tried to keep honest history and delete counterfeit inventory. That is a hard operation. It is also a political operation. Exchanges become unexpected referees. Bridge teams become veto players. Retail users become spectators who only see the token price and a status page. When a new exploit hits an EVM lending app months later, those spectators do not parse execution environments. They ask a blunter question. Is this place safe yet?

A fair answer is layered. A chain can repair a runtime bug and still host an application that mis-handles collateral. An application can fail without proving the base layer is broken. Both can be true on the same calendar day. Grown-up coverage has to hold both ideas at once.

Who Bears The Loss When A Reserve Goes To Zero

Noncustodial is a legal and technical claim. It is not a magic cloak over social obligation. If suppliers of WFLOW wake up to an empty reserve, they are the first loss tranche unless the team has insurance, a safety module, a treasury backstop, or a negotiated clawback. Borrowers who used the drained asset as the thing they borrowed may suddenly look healthy on paper while lenders look ruined. Liquidators may find there is nothing left to seize that can make lenders whole.

That distribution problem is why I watch for official pause announcements and for any hint of a reimbursement plan. Silence is not neutrality. Silence is just an unfinished sentence. Users should also watch whether related markets on the same protocol remain open. Contagion inside one lending deployment can jump from a single reserve to a confidence run across every listed market.

The Quiet Role Of Ratio Feeds And Reward-Bearing Shares

Reward-bearing tokens are elegant when they stay in a wallet. They get messy when a lending protocol needs a spot value every block. If value accrues in the exchange rate rather than in token quantity, the protocol must read that rate correctly, update it at the right times, and refuse to let an attacker force a stale or inflated reading through a crafted sequence. A ratio feed that is perfectly honest on a quiet Tuesday can still be a weapon on a hostile Monday if the consumer contract trusts it in the wrong order.

This is where I get a little opinionated. I would rather see staking derivatives isolated in conservative LTV buckets until a protocol has live battle time, not just a copied parameter table. Efficiency is addictive. It is also how a $9 million hole opens in an afternoon. Capital efficiency that depends on two assets remaining “close enough” is not free. Somebody is writing an implicit insurance policy. If no fund is reserved for that policy, lenders are the insurer and they never signed the form.

What A Serious Post-Mortem Needs To Answer

When the full write-up arrives, and it should arrive, I want more than a transaction hash collage. I want the entry point. I want the exact accounting mismatch or parameter abuse. I want to know whether a pause guardian existed and why it did or did not fire. I want the list of assets whose risk parameters will change. I want a statement on remaining solvent reserves. I want a clear sentence on whether Ankr contracts were merely used or actually flawed. Those are not niceties. They are how the next protocol on the same virtual machine avoids photocopying the same hole.

If a post-mortem cannot explain the drain in language a careful depositor can follow, it is a press note, not a security document.

Researchers already noted that more details were still being investigated after the 15.5 million WFLOW outflow and the later transfer cluster. That sentence should stay on the screen. It is the adult version of “we do not know everything yet.” The industry has a habit of filling silence with fan fiction. Resist that habit. The chain will still be there when the traces settle.

A Practical Framework For Judging Similar Markets

You do not need to become a professional auditor to get less naive. Look at whether a lending market lists a yield-bearing derivative next to its wrapped base asset in a high-efficiency mode. Look at who controls the ratio or price feed. Look at how new the listing is. Look at total value sitting in a single reserve. A concentrated reserve is a concentrated target. Fifteen million tokens in one pot is not a community piggy bank. It is a bounty poster.

Also look at operational maturity. Does the team have a documented incident channel? Is there a pause function with a known delay? Are risk parameters changed through a process that leaves a trail, or can a hurried listing sneak into production? These questions sound boring until they are the only questions that matter.

• Prefer markets that isolate staking derivatives from aggressive E Mode pairs.
• Treat brand-new listings as experimental even if the parent architecture is famous.
• Cap your deposit so a total reserve loss is painful, not existential.
• Assume detector impact and final loss will differ for at least a few days.

Sentiment, Listings, And The Memory Of The Last Crisis

Price action after an exploit is a rumor engine. Traders sell the token because they cannot sell the smart contract. That reflex is crude and sometimes correct. If users fear another recovery fight, another halt, another argument with venues, they do not wait for a white paper. They hit the bid. Flow already had to deal with listing pressure after the earlier token-duplication event. A fresh headline about a multi-million dollar drain, even if it is contained to one app, feeds that memory.

I do not think every application failure should be treated as a chain death sentence. I also do not think ecosystems get to advertise a lending market and a staking product, collect the usage, then act shocked when users connect the brands. Promotion creates association. Association creates responsibility in the court of user trust, even when the legal boundaries stay neat.

The Broader Lesson For Aave-Style Forks On Newer VMs

Forking a mature lending architecture is not the same as inheriting its scars. Ethereum’s major markets learned some lessons in public, at enormous cost. Copies on newer environments often launch with the features and without the institutional memory. E Mode gets turned on because the interface supports it. LSTs get listed because users ask for leverage on staked coins. Oracles get pointed at the most convenient feed. Then someone writes a contract that walks through that maze in an order no launch checklist imagined.

If there is a constructive way to read this incident, it is as a warning about feature completeness. Shipping every module from a reference design is not a badge. It can be a liability. Maybe the grown-up launch sequence is wrapped native first, conservative LTV second, staking derivatives later, efficiency modes last, and only after the pricing path has been abused in a public contest. That sequence is slower. Slower is less sexy. Slower also leaves fewer empty reserves.

What Remains Unknown, And Why That Unknown Is The Real News

As of the first public wave, we do not have a complete technical reconstruction. We do not have a confirmed final loss after slippage and partial recovery. We do not have a definitive statement that pins blame on a single line of code. We do not know whether funds will be frozen at any downstream venue. We do not know if More Markets can make suppliers whole. Those gaps are frustrating. They are also the honest state of the story.

What we do have is specific enough to act on. A named protocol. A named reserve. A size. A detected dollar impact. A described mechanism involving a bonded Ankr LST and E Mode. A published set of transactions. A reminder that Flow EVM applications can fail without the base chain repeating its Cadence nightmare. That package is already more useful than half the exploit threads that circulate with a screenshot and a theory.

I keep coming back to the human part. Somebody supplied WFLOW because a dashboard showed APY and a familiar interface. Somebody else staked FLOW through a liquid product because the pitch said you can keep using your capital in DeFi. Both actions are reasonable in a functioning market. Both become painful when the market’s internal model of “close enough” collateral turns out to be an opening. The code did not feel that pain. People did.

A Closing Read For Anyone Still Tempted To Shrug

Nine million dollars is not a rounding error, and it is not “just crypto being crypto.” It is a lending reserve that was supposed to stand behind suppliers. It is a design pattern that many other markets still copy. It is another data point that reward-bearing tokens and efficiency modes need tighter chaperones than launch blogs admit. If you want a single sentence to carry out of this, use this one. Correlation is not a security control.

Watch the transfer cluster. Watch the official post-mortem. Watch whether parameters change in public. And if you are still farming on a lookalike market somewhere else, open the risk page tonight, not after someone else publishes the next 15 million token screenshot. The next attacker is already reading the same disclosure you are. The only question is who turns the page first.