A bug that Cosmos Labs decided in May 2026 was not worth telling anyone about ended up draining six blockchains in the span of five days. The flaw, an integer underflow inside the shared Cosmos EVM codebase, let attackers turn a validator delegation into a balance that wrapped around to a 78-digit number, then used the same trick in reverse to siphon real tokens out of other people’s wallets. MANTRA, TAC, and KiiChain were hit within 45 hours of each other in August. A fourth chain, widely believed to be the AI-focused network Nesa, was drained four days later. Two more chains that Cosmos Labs has still not named round out what the firm itself now calls a $5.7 million loss event, though the face value of what actually moved on-chain runs well past $50 million once you add up every inflated balance.
What makes the incident notable isn’t the dollar figure, modest by 2026 DeFi-hack standards, but the timeline. A researcher flagged the underflow through Cosmos Labs’ bug bounty program back in April 2026. The team tested it, decided it could not be triggered on a live network, and rolled a quiet fix into its normal release cycle in May without telling any of the chains running its code. That judgment call is the real story of the Cosmos EVM vulnerability, and it is why validator teams across the ecosystem spent the last week of August rewriting their disclosure playbooks.
What Happened: A Bug Cosmos Labs Thought Was Dead
Cosmos EVM is the shared module that lets Cosmos SDK chains run Ethereum-compatible smart contracts. Dozens of independent networks, each with its own validator set, token, and governance, run the same underlying code rather than maintaining their own fork. That reuse is normally a strength: one audited codebase, patched once, protects an entire ecosystem. In August 2026 it became the opposite. A single accounting flaw in that shared module put every chain running an unpatched build at risk simultaneously, regardless of how careful any individual project’s own contract audits had been.
According to Cosmos Labs’ own post-mortem, filed under advisory GHSA-7g4w-cg88-2cq2, the bug was reported through the project’s bug bounty program in April 2026. Engineers tested it against production-like conditions, concluded live chains weren’t exploitable, and merged a fix in May as part of what the firm described as a routine, undisclosed patch. That practice is not unusual for the team: Cosmos Labs told The Block it has quietly patched 37 vulnerabilities this way over the preceding 13 months without issuing public advisories for any of them. Most of those, presumably, really were low-risk. This one wasn’t.
In early August, independent researchers determined the flaw was in fact exploitable on live networks. Cosmos Labs obscured a new fix to prevent reverse engineering and shipped it at 7:01 p.m. ET on August 19, with release notes referencing “important” security fixes but no specifics. Twenty hours later, at 3:06 p.m. ET on August 20, the first attack hit MANTRA.
Inside the Integer Underflow: How the Exploit Actually Worked
The mechanics are unusually clean for a bug that caused this much damage. An attacker set up an account holding locked or vesting tokens, then used the staking precompile, a bridge that exposes native Cosmos staking functions to EVM-style transactions, to delegate more tokens from that account than it actually held. Subtracting more than a balance contains should simply fail. Instead, the account’s balance ran past zero and wrapped around to the maximum possible value for a 256-bit unsigned integer: 2 to the power of 256, minus 1, a number 78 digits long.
The Setup
Underflow bugs are among the oldest classes of smart contract vulnerability, but they are supposed to have been solved years ago through checked arithmetic libraries. The Cosmos EVM flaw slipped past that protection because the vulnerable operation sat at the boundary between two accounting systems: the EVM-style balance ledger and the native Cosmos SDK x/bank module. Each side trusted the other’s math, and the underflow happened in a code path that neither side’s own checks fully covered.
The Reverse Trick
Once an attacker held an account showing a balance of roughly 78 digits, they ran the same arithmetic in the opposite direction. Sending that inflated balance to a target account pushed the target’s own balance past the same ceiling, this time overflowing rather than underflowing, and wrapping back down toward zero. The net effect: the attacker’s target account, usually a large, rarely-monitored wallet like a burn address or a dormant multisig, lost its tokens, and the attacker’s account gained them. Critically, no new tokens were minted in the process. Total token supply was, in a narrow technical sense, unchanged. What changed was who controlled it.
1. attacker delegates more locked tokens than account holds
via staking precompile
2. balance subtraction underflows: 0 - x wraps to 2^256 - 1
3. attacker "sends" the wrapped balance to a target account
(burn address, dormant multisig, large idle wallet)
4. target balance overflows past 2^256 - 1, wraps back near 0
5. result: target drained, attacker credited, total supply
unchanged on paper -- but control of real value has movedCosmos Labs said the targeted accounts were arbitrary large-balance wallets rather than any specific victim, which is why burn addresses and old multisigs from incentive campaigns ended up on the receiving end of the attacks. Nobody was actively using those wallets, which is also why nobody noticed for hours.
Timeline: From an April Bug Report to a Six-Chain Hack
Laid end to end, the sequence runs from a routine bug bounty submission to a multi-chain incident in about four months, with the actual damage compressed into less than a week:
- April 2026: An independent researcher reports the underflow through Cosmos Labs’ bug bounty program.
- May 2026: Cosmos Labs merges a fix silently, having judged the bug unexploitable on live chains.
- Early August 2026: Independent researchers determine the flaw does affect live Cosmos EVM chains.
- August 19, 7:01 p.m. ET: Cosmos Labs ships an obscured patch with vague release notes.
- August 20, 3:06 p.m. ET: First attack hits MANTRA, roughly 20 hours after the patch shipped.
- August 20, 7:13 p.m. ET: MANTRA halts its chain, resuming about 30 hours later on patched software.
- August 22: TAC and KiiChain are hit the same evening, roughly 45 hours after MANTRA. Cosmos Labs recommends all Cosmos EVM chains halt and upgrade.
- August 23: KiiChain publishes its own technical post-mortem, five days ahead of Cosmos Labs.
- August 24: Exchange Bitvavo suspends deposits and withdrawals for NES, Nesa’s token, citing a “critical consensus vulnerability.”
- August 26: Blockchain analytics firm Bubblemaps publishes analysis identifying Nesa as an affected chain.
- August 28: Cosmos Labs releases its official post-mortem under advisory GHSA-7g4w-cg88-2cq2.
- September 11: Rekt News publishes an independent reconstruction of the Nesa incident.
That 20-hour gap between the patch and the first attack is the detail every affected team has since complained about. MANTRA’s own post-mortem put it bluntly: “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,” the team wrote in its report. The release notes never said the fix was security-critical, so validator operators had no reason to treat it as urgent.
MANTRA: The Biggest Disclosed Hit
MANTRA absorbed the largest confirmed loss. The attacker moved roughly 600 million OM tokens out of the network’s burn address and another 120.9 million out of a dormant multisig left over from an old incentive campaign, a combined 720.9 million tokens the project valued at approximately $3.6 million using the pre-incident price. Because both source wallets had been excluded from circulating-supply calculations as unspendable, MANTRA’s reported circulating supply rose by that same 720.9 million once the tokens became tradable again.
MANTRA’s monitoring never flagged the first transaction. “The burn address is treated as immovable, so our monitoring systems did not cover transfers from it,” the team explained. The theft went undetected for nearly four hours while the attacker drained the dormant multisig too. The chain halted 14 minutes after a second unauthorized debit was finally caught, staying down for roughly 30 hours while validators upgraded to patched software without a rollback.
By the time the halt hit, the attacker had already moved 94.7% of the stolen OM, about 682.9 million tokens, to a single exchange deposit address across 15 separate transfers. Roughly 38 million tokens remained frozen in the attacker’s wallet. As of MANTRA’s August 28 update, none of the stolen funds had been recovered. OM’s price fell to an all-time low in the hours after the attack before rebounding about 14%, to roughly $0.00304, once the post-mortem was published and the network’s status stabilized.
TAC and KiiChain: Same Bug, Different Nights
TAC, a network bringing DeFi applications to TON and Telegram users, lost nearly 3 billion TAC tokens, roughly 2.99 billion by Rekt News’ reconstruction, drained from its staking pool on August 22. About 1.2 billion of those were sold on BNB Chain for approximately $950,000, a fraction of face value, since dumping billions of tokens into thin liquidity crushed the realized price. Rekt’s analysis notes TAC explored rewriting its own token-supply accounting to strand the attacker’s remaining holdings.
KiiChain, a network focused on foreign-exchange infrastructure, was hit the same evening, losing about 148.3 million KII tokens across 18 separate exploit transactions before validators halted the chain at block 9,355,723. Of that total, 64.6 million KII was sold for roughly $1.6 million. Cosmos Labs said it believes about 54% of the tokens taken from KiiChain remain recoverable on-chain if the network is restored to its pre-exploit state, though that would require coordinated validator action that hadn’t happened as of this writing.
KiiChain’s team was the most publicly critical of Cosmos Labs’ handling of the disclosure. Its own post-mortem, published August 23, five days ahead of Cosmos Labs’ official report, stated plainly: “Cosmos Labs gave no advance notice to downstream chains, did not flag the release as security critical, and did not tell affected chains that a public release had happened until Friday 21 August, two days later.” The team argued that speed mattered more than a polished patch: “A patch takes days to review, build, test and roll out across a validator set. A halt takes minutes.” KiiChain also disputed Cosmos Labs’ account of the fix itself, saying the exploit required three separate upstream defects rather than two, and that only the underflow issue had been patched publicly at the time of its report.
The Nesa Mystery: A $50 Million Headline, a $60,000 Reality
The fourth publicly identified chain tells a stranger story. Nesa, an AI-focused Cosmos EVM network, saw an attacker spend around $250,000 buying NES tokens, bridge them onto the network, exploit the same underflow bug to inflate the balance roughly 200-fold, and move 257,703,733 NES, close to 25.8% of the token’s total supply, back to Ethereum through the Hyperlane bridge. At NES’s prevailing price, that headline figure worked out to roughly $50 million.
The attacker never realized anywhere close to that. Blockchain analytics firm Bubblemaps, whose findings were later corroborated by Rekt News, found that dumping such a large position into limited liquidity produced extreme slippage. After liquidity providers pulled from the trading pools, the attacker walked away with roughly $60,000 in net profit against roughly $255,000 spent, a spectacular gap between the exploit’s face value and its real payout. Bubblemaps also noted the attacker’s wallet was originally funded through Monero and that the funding pattern differed enough from the MANTRA, TAC, and KiiChain attacks that a separate party may have carried out the Nesa exploit independently, copying the disclosed technique rather than coordinating with whoever hit the first three chains.
Exchange Bitvavo suspended NES deposits and withdrawals on August 24, citing a critical consensus vulnerability that let compromised nodes accept invalid blocks. Nesa itself has not published its own post-mortem, and Cosmos Labs’ official report never names the project directly, referring only to unnamed additional affected chains.
Two Chains Still Unnamed
Cosmos Labs’ post-mortem states six chains total were hit using the same method, but the firm has only confirmed MANTRA, TAC, and KiiChain by name. Nesa’s involvement rests on exchange action and independent analyst reporting, not an admission from Nesa or Cosmos Labs. Two chains have never been made public, an unusual gap for an incident this well documented elsewhere. Neither Cosmos Labs nor Nesa responded to press requests for comment, according to The Block.
Cosmos Labs’ Silent Patch Process Under Fire
The central tension in this incident isn’t the underflow bug itself, underflow and overflow errors are a known, well-studied class of vulnerability, it’s the decision to patch quietly rather than warn downstream operators. Cosmos Labs’ defense is straightforward: broadcasting the details of an unpatched vulnerability before every chain can apply a fix hands attackers a roadmap. Silent patching, shipped inside otherwise routine releases, is meant to close the gap before anyone notices it existed.
That approach depends on getting the initial risk assessment right, and this time Cosmos Labs got it wrong. Testing that concludes a bug isn’t exploitable, then reversing that conclusion months later, is exactly the scenario silent patching can’t handle: the fix sat public in the commit history for weeks before anyone flagged the real risk. MANTRA has since formally raised the delay with Cosmos maintainers, asking for “clearer disclosure practices and defined backport expectations for security-relevant fixes.” KiiChain argued a same-day halt recommendation, not a quiet patch with vague release notes, was the only measure that could have contained the damage once exploitability was confirmed.
Market Impact: Token Prices, Exchange Response, and Investor Confidence
None of the four confirmed tokens suffered a catastrophic, FTX-style collapse, but each took a real hit. OM fell to an all-time low before its partial rebound. NES faced an exchange-level trading suspension. TAC and KII both had to explain sudden, unexplained multi-hundred-million-token movements to their own communities before any official cause was confirmed. The table below summarizes what’s publicly known about each chain’s exposure.
| Chain | Date Hit | Tokens Moved | Reported USD Value | Recovery Status (as of report) |
|---|---|---|---|---|
| MANTRA (OM) | Aug 20, 2026 | 720.9 million OM | ~$3.6 million | 0% recovered; ~38M frozen at halt |
| TAC | Aug 22, 2026 | ~2.99 billion TAC | ~$950,000 realized (1.2B sold) | Supply-rewrite remedy explored |
| KiiChain (KII) | Aug 22, 2026 | 148.3 million KII | ~$1.6 million realized (64.6M sold) | ~54% estimated recoverable on-chain |
| Nesa (NES, unconfirmed) | Aug 24, 2026 | 257.7 million NES | ~$50M face value; ~$60,000 net profit | Not officially acknowledged by Nesa |
| Two unnamed chains | Unknown | Not disclosed | Included in $5.7M aggregate | Never publicly identified |
Cosmos Labs’ own $5.7 million aggregate figure reflects realized or converted value across all six chains, not the sum of every inflated balance at spot price. That distinction matters for anyone trying to size the actual damage: Nesa’s headline $50 million evaporated almost entirely to slippage, while MANTRA’s $3.6 million loss was close to fully realized because the attacker moved it through an exchange before the chain halted. No major exchange has confirmed a permanent delisting tied to the incident, and no regulator has announced a formal enforcement action as of this report. The incident adds to a year that already set records for hack frequency, even as aggregate dollar losses across the industry have trended down from their 2025 peaks.
How This Compares to Past Cross-Chain and Bridge Hacks
Shared-infrastructure bugs that hit multiple independent chains at once are rarer than single-protocol exploits, but they aren’t new. The table below places the Cosmos EVM incident alongside several of the best-documented bridge and cross-chain failures of the past few years.
| Incident | Year | Approx. Loss | Root Cause |
|---|---|---|---|
| Ronin Bridge (Axie Infinity) | 2022 | ~$625 million | Validator private keys compromised |
| Poly Network | 2021 | ~$611 million | Cross-chain contract privilege abuse |
| Wormhole | 2022 | ~$325 million | Signature verification flaw |
| Nomad Bridge | 2022 | ~$190 million | Message-validation failure, mass-copied exploit |
| Cosmos EVM (six chains) | 2026 | ~$5.7 million realized | Shared-module integer underflow/overflow |
By dollar value, the Cosmos EVM incident is small next to Ronin, Poly Network, or Wormhole. What sets it apart is breadth per line of vulnerable code. Ronin, Wormhole, and Nomad were each single-bridge failures. One patch to one shared module put six independent chains at simultaneous risk, a blast radius tied less to the severity of the bug than to how much infrastructure now depends on the same codebase. It’s the same structural risk researchers flagged around Liquid Network’s federation model, where a consensus-validation bug, not a stolen key, moved roughly $320 million out of Blockstream’s Bitcoin sidechain in September.
Why Shared Infrastructure Bugs Are More Dangerous Than Single-Protocol Exploits
Application-level DeFi exploits, the price-oracle manipulations, flash-loan attacks, and reentrancy bugs that make up most of the industry’s hack volume, are contained by definition. A flawed lending market can only drain the money inside that lending market. A bug in shared base-layer infrastructure doesn’t respect those boundaries. Every chain that reused the vulnerable Cosmos EVM module inherited the exact same risk the moment the code was merged, regardless of how well-audited each individual chain’s own application contracts were, a gap standard smart contract audit processes aren’t designed to catch since the vulnerable code lives one layer below the application itself.
That’s also why the disclosure question matters more here than in a typical single-protocol hack. When one team finds a bug in its own contract, it can patch and redeploy on its own timeline. When the bug lives in shared infrastructure serving dozens of independent validator sets, no single team controls the rollout schedule. Cosmos Labs could ship a fix, but it could not force MANTRA, TAC, KiiChain, or Nesa to treat an unlabeled release as urgent. That coordination gap, not the underflow arithmetic itself, is what turned a single bug report into a six-chain incident.
What Cosmos Labs, MANTRA, and KiiChain Are Doing Differently Now
In the wake of the incident, the affected teams have pushed for the same basic fix from different angles. MANTRA wants Cosmos Labs to commit to vulnerability-specific advisories for anything that could break state, rather than folding security fixes into generic release notes. KiiChain argues halt recommendations should go out the moment exploitability is confirmed, since a halt takes minutes while a coordinated upgrade takes days. Cosmos Labs has acknowledged it misjudged the April report and says it is revising its triage process, though it hasn’t published details or a timeline.
KiiChain’s separate claim, that the exploit chain actually involved three distinct upstream defects rather than the two Cosmos Labs described, remains unresolved in the public record. If accurate, it would mean part of the vulnerable surface was still unpatched even after the August 19 release, an uncomfortable possibility for any chain that assumed upgrading to the patched build fully closed the door.
What This Means for Developers Building on Cosmos EVM
For teams building on Cosmos EVM chains, the lesson isn’t to audit application code more carefully, most affected projects likely had reasonably audited application layers already. It’s to track dependency risk in the base layer itself. A shared-module patch note reading only “important security fixes” deserves investigation, not a skim. Monitoring built on the assumption that certain wallets, burn addresses, old multisigs, treasury reserves, can’t move funds is exactly the blind spot this exploit hit; MANTRA’s own admission that its burn address sat outside transfer monitoring is worth checking against any chain holding large, rarely-touched balances.
Teams running their own validator infrastructure on Cosmos EVM chains should also treat “important security fixes” language in release notes as reason enough to expedite an upgrade, rather than waiting for an official advisory that, in this case, arrived roughly 20 hours too late for three separate networks. Security teams that already run bridge exploit testing against their own contracts should extend that same discipline to shared modules they don’t maintain themselves.
Predictions: Where This Goes From Here
- Cosmos Labs will likely publish a formal disclosure-policy update within the next quarter, given both MANTRA’s and KiiChain’s public pressure for vulnerability-specific advisories.
- Expect at least one of the two still-unnamed chains to surface publicly once on-chain sleuths finish cross-referencing wallet-funding patterns against the known attack windows.
- KiiChain’s claim of a third unpatched upstream defect will probably force a second Cosmos EVM security release before the end of 2026, whether or not it’s formally linked to this incident.
- Other multi-chain SDK ecosystems, Polkadot’s parachains, Avalanche subnets, will face renewed scrutiny over how they disclose shared-module vulnerabilities to downstream, independently operated chains.
- MANTRA’s stolen funds, still sitting mostly on a single exchange deposit address, carry a reasonable chance of at least partial freezing if that exchange cooperates with law enforcement, following the pattern set by several 2025 and 2026 recovery cases.
The Bottom Line
The Cosmos EVM incident is a reminder that code reuse across independent blockchains carries the same tradeoff it carries in traditional software supply chains: one shared library means one shared blast radius. A bug judged low-risk in April, patched quietly in May, and left undisclosed through August cost six networks a combined $5.7 million and, more expensively, months of trust rebuilding between Cosmos Labs and the ecosystem of chains that depend on its code without controlling its release calendar. The dollar figure here is small by the standards of 2026’s biggest DeFi exploits. The coordination failure behind it is not.
Frequently Asked Questions
What is the Cosmos EVM vulnerability, in plain terms?
It’s an integer underflow bug in the shared Cosmos EVM module that let an attacker make an account’s balance wrap around to a near-infinite number, then run the same math in reverse to drain a real, unrelated wallet’s tokens without minting any new supply.
Which chains were confirmed affected?
Cosmos Labs has officially named MANTRA, TAC, and KiiChain. Nesa is widely believed to be a fourth affected chain based on exchange action and independent analyst reporting, though neither Nesa nor Cosmos Labs has confirmed it by name. Two additional chains referenced in Cosmos Labs’ post-mortem have never been publicly identified.
How much money was actually lost?
Cosmos Labs puts the aggregate realized or converted value across all six chains at approximately $5.7 million. Individual headline figures, like Nesa’s roughly $50 million in nominal token value, are much higher on paper but were sharply reduced by slippage when attackers tried to liquidate the positions.
Was this a case of stolen private keys?
No. No validator keys, multisig signatures, or wallet credentials were compromised. The exploit relied entirely on an accounting flaw that let the network itself miscalculate balances, a fundamentally different and, in some ways, harder-to-defend-against failure mode than a conventional key theft.
Why did Cosmos Labs patch the bug quietly instead of issuing a public warning?
Cosmos Labs’ standard practice is to patch suspected low-risk bugs silently, folding fixes into routine releases without advisories, to avoid handing attackers a roadmap before every downstream chain can upgrade. The firm says it has patched 37 vulnerabilities this way over 13 months. The approach failed here because the initial risk assessment concluded the bug wasn’t exploitable on live chains, a conclusion reversed only after independent researchers proved otherwise in early August.
Have any of the stolen funds been recovered?
As of the most recent public updates, MANTRA reported zero recovery, with about 38 million of its stolen OM frozen in the attacker’s wallet when the chain halted and the rest already moved to an exchange. Cosmos Labs estimated roughly 54% of KiiChain’s stolen KII remains recoverable on-chain if the network coordinates a restoration. No comprehensive recovery has been confirmed for TAC or Nesa.
Is Cosmos EVM safe to build on now?
The specific underflow/overflow path has been patched as of the August 19 release, and MANTRA’s post-mortem describes that fix as closing the attack path it verified. KiiChain disputes that the fix is complete, claiming a third upstream defect remains unpatched. Developers and validator operators should treat “important security fixes” language in future Cosmos EVM release notes as a signal to upgrade immediately rather than wait for a formal advisory.
How does this compare to other cross-chain and bridge hacks?
In dollar terms it’s far smaller than incidents like Ronin ($625 million), Poly Network ($611 million), or Wormhole ($325 million). What makes it structurally significant is that one vulnerability in shared infrastructure hit six independent, sovereign chains almost simultaneously, rather than a single bridge or protocol, underscoring how much of the multi-chain ecosystem now depends on the same underlying code.
