MultiversX Suspends Mainnet Due to Virtual Machine Atomicity Vulnerability
MultiversX has confirmed that attackers attempted to exploit an atomicity flaw in the mainnet virtual machine layer, resulting in invalid state changes on the chain. To prevent further spread, the network has paused block production. Previously, the official statement only mentioned an investigation into potential mainnet issues, which has now escalated to a shutdown and repair process.
The engineering team has written a patch, which will first be validated in a shadow fork environment; after successful testing, coordination will occur with validation nodes, exchanges, and infrastructure providers to deploy the patch on the mainnet. A parallel assessment for targeted recovery is underway: preserving confirmed transaction history and legitimate user states while only rolling back or correcting invalid changes related to this incident. A complete technical report will be released after the event concludes and investigation results are confirmed.
The official request to users is not to operate independently: do not submit or replay transactions until recovery notifications are issued, and do not deposit or withdraw egld and esdt through exchanges or cross-chain bridges. This is because if the state machine contains invalid changes, off-chain deposits and bridging will replicate incorrect balances to custodial accounts. On September 10, the network completed the Supernova hard fork, which had a routine downtime of about 24 minutes; this incident is a security response, and no end time for the downtime has been provided.
This is not the first time similar issues have affected the same virtual machine. In May 2025, a transaction from a multi-signature contract exploited a lack of verification in the transferValueAndExecute opcode, writing esdt balances to the target shard, effectively unauthorized minting. At that time, the team temporarily shut down with over 25 staking providers to apply a patch. This time, the description has shifted from "lack of verification in opcode" to "atomicity," pointing to the same execution layer: when crossing shards or calls, success and failure are not bound to the same submission.
The market mechanism is that liquidity is first shut off by the gate. Sellers are holders and market makers unable to withdraw; buyers are short positions or funds pricing in advance on the opposite side of the bridge during the downtime window. Exchanges and cross-chain bridges typically freeze egld and esdt channels; spot markets remain, but on-chain settlements have disappeared. Beneficiaries are those able to participate in shadow fork validation and grasp recovery heights; those under pressure are applications relying on instant finality, staking redemptions, and cross-chain inventories. Funds are temporarily held in custodial hot wallets and bridge contracts, waiting for the patch to determine which balances are valid.
Source: Public Information
ABAB AI Insight
Since rebranding from Elrond, MultiversX's selling point has been high throughput through sharding and self-developed virtual machines. Sharding splits a single call into multiple execution environments, and atomicity must be guaranteed by the protocol rather than the contract authors. The previous esdt inflation incident in 2025 has already proven that cross-shard receipts can add balances in the target shard without actual deductions in the source shard. Discussing atomicity again indicates that the submission boundaries of the execution layer are still not closed. The shadow fork replicates the production state to an isolated network to validate the patch, avoiding direct block production on the mainnet during the next round.
The capital path follows the custodial flow. A large amount of egld and esdt is stuck in exchanges and bridges; the shutdown order temporarily recovers issuance rights to coordinate with core developers and validators. The commitment to targeted recovery of "only cutting invalid changes" seeks a compromise between immutable narratives and necessary accounting adjustments: if adjusted less, attack balances remain; if adjusted more, it equals a social consensus fork. The Supernova consensus parameters were just adjusted, and now due to a security shutdown, validators will need to run two sets of binaries, and how staking rewards and penalty rules are calculated during the empty block period will become the next governance topic.
Analogies can be drawn to Solana's multiple downtimes, BNB Chain's post-bridge state adjustments, Ronin's ledger negotiations after theft, and Harmony's hard fork debates post-door theft. Once a public chain writes "final confirmation" as a rollable subset, finality degrades from cryptographic commitment to operational processes. The industry phase is exposing control layers: the narrative of throughput has expanded for years, yet execution atomicity remains a single point of failure.
Structurally, this belongs to the reconstruction of the industrial chain after the failure of technical alternatives. The mechanism is: sharding increases parallelism but splits the correctness of a transaction across multiple state machines; attackers buy into a "half-success" intermediate state. Whoever can decide which intermediate states are invalid temporarily holds the pricing power of the ledger. Users are asked to sit tight because off-chain channels replicate errors faster than on-chain.
ABAB News · Cognitive Laws
Sharding accelerates parallelism but also accelerates half-success states
Finality that can be selectively rolled back is no longer finality
Bridges and exchanges replicate erroneous balances earlier than blocks