MultiversX paused block production after an attacker targeted a VM-layer atomicity flaw The co-founder says the incident is contained and the attacker’s accounts have been frozen with exchang
MultiversX said the attack which forced it to pause its mainnet is now contained, with the attacker’s accounts identified and frozen through coordination with major exchanges and a recovery patch expected to run within one to two days. The chain remains halted while engineers validate the fix in a shadow fork environment. The underlying fault was an atomicity bug at the virtual machine layer, a flaw that let invalid changes settle onto the ledger before the team stopped block production to contain it. Users were told to sit still: no new transactions, no resubmitting old ones, and no EGLD or ESDT deposits and withdrawals through exchanges or bridges until the all-clear.
Update:
We have confirmed that a actor attempted to exploit a VM-level atomicity issue.
The attempt caused invalid state changes, and network progression remains paused to prevent any further impact. Engineering has prepared a fix and will validate it on a shadow fork.… https://t.co/iyi1aYzkMJ
— Multiversᕽ (@MultiversX) September 19, 2026
Atomicity is one of the load-bearing guarantees in any transaction system. A transaction with several internal steps has to either complete in full or fail in full, with nothing left half-done. When you send tokens, one balance goes down and another goes up in the same indivisible operation. There is no valid state in between.
The attacker found a way to break that guarantee at the VM level. By exploiting the atomicity handling, they forced the network to record invalid state changes, the kind of outcome where a credit lands without its matching debit, or a step commits that should have reverted. On a chain that runs parallel processing lanes across shards, a defect this deep is more dangerous than an ordinary bug, because the corrupted state can propagate before anyone isolates it. That is the reason the response was a full stop rather than a quiet patch.
Stopping block production looks alarming, but here it functioned as a circuit breaker. Halting the chain stops new transactions from stacking on top of records that may already be wrong, and it blocks any repeat attempt using the same method while engineers map which balances and contract entries were touched. The freeze does not reverse anything. Every record the network already accepted stays in place as the starting point for wallets, applications and bridges until a recovery plan is adopted.
The damage did spread to the surface layer. As of publication, MultiversX marked its system as partially degraded, with the public API, xPortal, Explorer, Wallet, Bridge, xExchange and xLaunchpad all showing degraded performance, while the gateway and index stayed operational.
A shadow fork is a private, running copy of the live network. Engineers replay real mainnet history against the new code there, which lets them confirm the patch closes the atomicity hole without risking the actual ledger. Only after that validation passes does the fix reach production. MultiversX said it would then coordinate the deployment with validators, exchanges and infrastructure partners rather than pushing it live unilaterally, since a chain restart only works if node operators and custodians move together.
The clearest new detail came from co-founder Beniamin Mincu, who said the attacker’s accounts had been identified, seized and frozen in coordination with major global exchanges, and that specialized law enforcement was now involved. He described the incident as contained and user funds as safe, and for the first time put a clock on the repair: the recovery path is being tested and, once it clears, will run over the next one to two days. These are the team’s statements rather than independently verified facts, and MultiversX has said a full technical incident report will follow once the response is complete.
Solid progress made securing the situation.
1. The incident is contained.Fund are safe. In coordination with major global exchanges, the attacker’s accounts have been identified, seized and frozen. Specialized law enforcement is now engaged, and the attacker will face the… https://t.co/WDy2wnsHOu
— Beniamin Mincu |
/acc (@beniaminmincu) September 20, 2026
The harder question is not the patch. It is how to clean the ledger without punishing everyone who transacted during the attack window. The team said it is evaluating a targeted recovery that preserves finalized transaction history and legitimate user state, handling only the invalid changes tied to the incident. That is the opposite of a blunt rewind.
Full rollbackWhat it does. Rewinds the chain to a block before the attack and rebuilds from there.
Cost to users. Legitimate transfers made after that point can vanish and need repeating.
Targeted purgeWhat it does. Removes only the specific invalid state created by the exploit.
Cost to users. Keeps finalized legitimate history, but must prove it caught every bad change and nothing else.
The difficulty with the surgical option is verification: the team has to demonstrate that the correction captures every invalid change and touches no legitimate one. Web3 purists will read this as a test of whether “immutability” bends when the recorded state is provably corrupt. In practice, most large proof-of-stake networks have already answered that question by pausing and correcting when the alternative was letting bad state compound.
EGLD rebounded from its halt-day low near $3.50 to about $3.95. Source: Binance via TradingView.
EGLD did not keep falling. The token bottomed near $3.50 during the September 20 halt, then turned higher as the team signaled the situation was under control. After Mincu said funds were safe and the attacker’s accounts had been frozen, EGLD ran as high as $4.28 before sellers faded the spike, and it now trades around $3.95, with the latest four-hour candle down about 4.8% from that intraday top.
The moving averages describe a market that has gone flat rather than trending. The 20-period average sits at $3.956, the 50-period at $3.993 and the 200-period at $3.997, all bunched inside a band a few cents wide, with price resting right on top of them. When these averages compress into one zone, recent buyers and sellers have reached rough parity, and the token needs a decisive close above or below the band to pick a direction. Short-term momentum has ticked up: the MACD histogram has flipped positive and its faster line has crossed back above the slower one, an early sign that downward pressure is easing. Both lines still sit under the zero mark, so this reads as a bounce inside a soft trend, not a confirmed reversal. The rally and the fade both printed on heavy volume, which points to active two-way positioning around the news rather than a thin drift.
Context matters here, because the network was in the middle of its most active upgrade cycle in years. Supernova went live on September 10 and decoupled consensus from execution to speed up block times. That upgrade was approved in an on-chain vote in January 2026 with 99.64% support, stress-tested through March, and cleared an external security audit in June. Days later the team shipped a separate emergency patch. A native delegation accounting bug had corrupted staking rewards, and the fix that restored reward recording and claiming activated at epoch 2238, with earlier overwritten rewards set aside for a later reconstruction patch.
Sep 10, 2026 Supernova hard fork goes live Consensus separated from execution, round time cut below six seconds. Sep 15, 2026 Emergency staking-rewards patch Delegation accounting bug fixed at epoch 2238; claiming reopened. Sep 19, 2026 Mainnet anomaly detected Labeled a potential issue, with a full update promised within 12 hours. Sep 20, 2026 Exploit confirmed, chain paused VM atomicity flaw targeted; block production stopped to freeze state. Sep 21, 2026 Containment update, recovery in testing Co-founder says attacker accounts frozen and law enforcement engaged; fix to run within one to two days.The initial September 19 disclosure was deliberately narrow, reporting no stolen funds, compromised wallets or financial losses, and the exploit label only came a day later once the team had traced the behavior.
The next pressure point is the restart itself. With blocks frozen, every DeFi position on MultiversX is suspended in time, including loans, leveraged trades and automated market maker pools. When the chain comes back, price feeds that update on resumption can jump straight to a level well away from where they last printed, and any lending protocol that reads those feeds could fire liquidations the moment it reconnects. Leveraged borrowers who cannot add collateral during a halt are the ones exposed to that gap. How each protocol handles the first blocks after restart, whether it delays liquidations or eases feeds back in, will decide whether the reopening is orderly or produces a fresh wave of forced selling.
For a reference on timing, BNB Chain paused its network in October 2022 after a cross-chain bridge exploit and restarted within hours once validators coordinated a hot fix and a state correction, a close analogue to the targeted purge MultiversX is now weighing. The team says funds are safe and the attacker’s accounts are frozen, yet it has not published a figure for how much invalid state the exploit produced, and an external audit cleared Supernova in June, yet the attacker still found a viable VM path. That gap is what the promised incident report will have to close.
The post MultiversX Says Funds Safe, Attacker Accounts Frozen After Exploit appeared first on ETHNews.
