The Shared Module as a Single Point of Failure: A Cosmos EVM Flaw Hit Six Chains and Roughly USD 5.72 Million Was Cashed Out

Six chains in five days, roughly USD 5.72 million converted

The attacks ran from 19:06 UTC on August 20 to 15:20 UTC on August 25. According to Cosmos Labs, six networks were actually exploited. Approximately USD 2.87 million (approx. JPY 460 million) of the stolen assets was bridged to other chains and swapped on decentralized exchanges, and an estimated further USD 2.85 million (approx. JPY 460 million) was sold on centralized exchanges. The DEX figure breaks down into 2,613,674.48 USDT, 114.129045 ETH, 93.78 TON, 1.393618 USDC and 98.87 OSMO, based on August 19 prices; Cosmos Labs states explicitly that these numbers were supplied by the affected chains and have not been independently audited. The centralized exchange accounts used by the attackers had been frozen pending investigation by the relevant authorities as of publication.

Three of the affected chains have been publicly named: MANTRA, TAC and KiiChain. MANTRA lost 720,923,967.99 tokens from two addresses it controlled, worth roughly USD 3.6 million (approx. JPY 580 million) at the pre-incident price, and halted its network for about 30 hours and 13 minutes. On TAC, roughly 2.986 billion TAC was withdrawn from the bonded tokens pool and bridged to BNB Chain via LayerZero, after which about 1.2085 billion TAC was sold for 950,293 USDT across roughly 80 transactions. On KiiChain, approximately 148.3 million KII was drained through 18 repeated iterations, though about 54.4 percent of it remained on the network thanks to the chain halt and is expected to be recoverable after restoration.

The technical core: two ledgers disagreed on what a balance is

At the heart of the flaw was an inconsistency in how Cosmos EVM and the Cosmos SDK accounted for token balances. The Cosmos EVM StateDB tracks only an account's spendable balance, whereas vesting accounts in SDK state hold both a spendable and a locked balance, and both the staking module and the staking precompile permit delegating the locked portion. When an account delegated more than its spendable balance, the write-back subtraction underflowed and wrapped the balance to approximately 2 to the power of 256. The attacker then sent that impossible balance to a high-balance account such as the zero address or a genesis-era multisignature wallet, triggering an overflow that left the victim's legitimate tokens in the attacker's hands. The two defects were chained inside a single transaction and were designed to be supply-neutral. Because no new tokens were minted, monitoring total supply alone would not have surfaced the attack.

Two miscalculations: a wrong scope call and a broken silent patch

The first miscalculation concerned scope. The initial report, received through the bug bounty program on April 25, included a proof of concept on a network configured with six decimal places. Cosmos Labs could not reproduce the issue on 18-decimal configurations and concluded that no live production chain was affected. Classified as not causing fund loss, the fix was merged quietly into the public main branch on May 15 under the company's silent patch process. In early August, additional reports from independent researchers established that all Cosmos EVM chains were affected regardless of decimal configuration. Because the fix was already public and no exploitation had been observed, the company continued with the silent patch route rather than switching to private distribution, publishing patched releases v0.6.2 and v0.7.2 on August 19 with release notes that mentioned security but conveyed neither urgency nor severity.

The second miscalculation was external. At 07:16 UTC on August 20, a public pull request in Push Chain's fork of Cosmos EVM described the vulnerability and its exploitation path in detail, attributing the finding to an audit by security firm Hacken. The version table in that pull request omitted v0.6.2 and v0.7.2, published roughly eight hours earlier. Approximately twelve hours later, the first attack landed on MANTRA. The post-mortem does not assert causation, but Cosmos Labs notes that it has silently patched 37 vulnerabilities over the past thirteen months without a downstream developer publicly describing an exploit path in this way.

Objections from the affected chains, and an ecosystem without a registry

The affected teams do not fully accept the account. In its own post-mortem, MANTRA said that twenty hours was not a realistic window in which to assess, build, test and coordinate a state-breaking upgrade across 38 independent validators, particularly without a vulnerability-specific advisory, and that it has formally raised the matter with the Cosmos maintainers seeking clearer disclosure practices and defined backport expectations. KiiChain went further, arguing that only one of three defects, the underflow, has been fixed upstream, and that chains applying the official patch alone remain exposed to the other two.

Cosmos Labs also acknowledges a structural weakness of its own. The Cosmos ecosystem spans more than 115 known public blockchains, and because the software is open source and permissionless to deploy, the company holds no complete registry of who is running it. During this response it discovered eleven Cosmos EVM deployments that had never registered with its security communication channels. It coordinated with forty Cosmos chains and worked with thirteen potentially exposed networks to patch, halt or otherwise mitigate without further incident. Going forward, the company intends to overhaul its triage process for critical vulnerabilities, publish clearer standards for when a network halt is recommended over a coordinated upgrade and when private channels replace the public silent patch process, and commission an external audit of its overall security operations. Notably, the Cosmos EVM codebase had been audited by Sherlock in July 2025, and this vulnerability was not discovered in that review.

[Business Development Insights]

1. The practical unit of risk is not which chain you choose, but what that chain is running underneath. None of MANTRA, TAC or KiiChain had a defect in its own code; a single flaw in a shared upstream module broke all three within days. Financial institutions selecting a chain for tokenized assets or stablecoin issuance should evaluate not only the chain itself but the version of its base software, who maintains that upstream code, the patch application SLA, and how long a state-breaking upgrade takes to coordinate. There is a clear opening to formalize a software-bill-of-materials equivalent as a standard item in blockchain selection due diligence.
2. A chain halt is an availability risk that maps directly onto settlement and securities-servicing SLAs. MANTRA stopped producing blocks for about 30 hours and KiiChain for more than a week. A halt preserves user balances, but during it neither settlement nor redemption can be executed. MANTRA holds a Dubai VARA license, counts Securitize among its active validators, and hosts a stablecoin it describes as backed by short-term US Treasuries. Institutions structuring tokenized products need to write alternative execution paths, halt-decision authority, and redemption grace provisions into contracts and operating designs in advance.
3. Registration with upstream security notification channels is itself an asset that determines whether you can defend at all. Eleven deployments unknown to Cosmos Labs surfaced during this incident, and private notifications reached only those teams that had already supplied security contacts. In the post-mortem the company explicitly invites not just developers but security personnel at major bridges, exchanges and off-ramps to join. For any firm adopting a blockchain stack, registering with the upstream coordinated vulnerability disclosure channel, completing the associated KYC, and maintaining contact responsiveness is a zero-cost first line of defense that belongs in documented internal operating procedure.

[Sources]

New Economy, "Cosmos Labs Publishes Post-Mortem on Cosmos EVM Attack; Vulnerability Exploited on Six Chains" https://www.neweconomy.jp/posts/603746

Cosmos Security, "Cosmos EVM GHSA-7g4w-cg88-2cq2 Post-Mortem" (August 28, 2026) https://github.com/cosmos/security/blob/main/communications/cosmos_evm_GHSA-7g4w-cg88-2cq2_post_mortem.md

GitHub Security Advisory, "Balance underflow in EVM StateDB," GHSA-7g4w-cg88-2cq2 https://github.com/cosmos/evm/security/advisories/GHSA-7g4w-cg88-2cq2

pushchain/push-chain-evm Pull Request #40 https://github.com/pushchain/push-chain-evm/pull/40

The Block, "Cosmos Labs says it wrongly cleared the bug behind a $5.7 million six-chain hack" https://www.theblock.co/news/defi/2026-08-29-cosmos-labs-says-it-wrongly-cleared-the-bug-behind-a-5-7-million-six-chain-hack-413061

CryptoSlate, "Cosmos misjudged a critical bug for 4 months before hackers stole nearly $6 million across 6 chains" https://cryptoslate.com/cosmos-misjudged-a-critical-bug-for-4-months-before-hackers-stole-nearly-6-million-across-6-chains/

The Defiant, "Cosmos Labs Urges EVM Chains To Halt As Shared Bug Drains Three Networks" https://thedefiant.io/news/blockchains/cosmos-labs-urges-evm-chains-halt-shared-bug-drains-three-networks

The Defiant, "MANTRA Halts Chain, Blames Cosmos EVM Module" https://thedefiant.io/news/blockchains/mantra-halts-chain-blames-cosmos-evm-module

The Hacker News, "Cosmos EVM Flaw Exploited After Cosmos Labs Knew Every Blockchain Running It Was Vulnerable" https://thehackernews.com/2026/08/cosmos-evm-flaw-exploited-after-cosmos.html

Crypto Times, "Cosmos Labs Urges Cosmos EVM Chains to Halt Validators, Three Now Offline" https://www.cryptotimes.io/2026/08/25/cosmos-labs-urges-evm-chains-to-halt-amid-security-incident/

MANTRA Chain, "Securitize Joins the MANTRA Chain Validator Set" https://mantrachain.io/resources/announcements/securitize-joins-the-mantra-chain-validator-set