Cosmos Labs has disclosed that attackers exploited a critical vulnerability in Cosmos EVM across six blockchain networks between August 20 and August 25, converting stolen tokens into approximately 5.72 million USD in other assets through decentralized and centralized exchanges. The post-mortem, published by Cosmos Labs, reveals a timeline in which a patch existed for months before the attacks, but network operators were never specifically warned.
The disclosure has ignited one of the sharpest governance debates in the Cosmos ecosystem since the rise of cross-chain infrastructure, because the flaw was known, fixed, and silently shipped before anyone was told what it fixed.
## The vulnerability explained
The bug was an integer underflow in Cosmos EVM, the ecosystem’s Ethereum-compatible framework built from the open-source Evmos codebase. The exploit sequence sounds simple in retrospect. An attacker first created an account containing locked tokens, then delegated more tokens to a validator than the account was able to spend. Subtracting the delegated amount pushed the balance below zero, causing the value to wrap around to the maximum possible figure of 2^256-1 base units.
With an inflated balance in hand, the attacker targeted large accounts, including burn addresses and multisignature wallets created when networks launched. Sending the wrapped amount to a target pushed its recorded balance past the same numerical ceiling, causing an overflow that wrapped the value back down and left the attacker holding the target’s tokens.
Cosmos Labs emphasized that no additional tokens were created through the process and total token supply remained effectively unchanged. MANTRA, the hardest-hit network, said the exploit altered its supply by only a single base unit, the smallest divisible denomination of the token. The theft was a transfer, not a mint, which is why account selection mattered: attackers went after addresses holding large balances that could not defend themselves.
## A patch without a warning
The most consequential detail in the disclosure concerns timing. The vulnerability was first reported through the Cosmos bug bounty program on April 25, nearly four months before the attacks began. Cosmos Labs testers were unable to reproduce the exploit against configurations used by known production Cosmos EVM networks and concluded that live user funds were not at risk.
Based on that assessment, developers handled the fix through a silent public patch, merging it in May without telling network operators which vulnerability it addressed. The assessment collapsed in early August, when independent researchers established that the bug affected all Cosmos EVM chains. Cosmos Labs then obscured the fix to make reverse engineering more difficult and released patched versions at 7:01 PM ET on August 19. The release notes referenced important security fixes without describing the vulnerability.
The first known attack began at 3:06 PM ET on August 20, roughly 20 hours after the patched software became available. MANTRA later argued that 20 hours was not enough time to assess, build, test, and coordinate a state-breaking upgrade across its validator set, a criticism that has resonated across the ecosystem.
## The victims
MANTRA suffered the largest single loss, with 720.9 million tokens worth approximately 3.6 million USD drained. TAC and KiiChain were subsequently attacked using the same method, and the total number of affected networks reached six. Cosmos Labs said it coordinated with 40 chains during the response and helped 13 networks patch or halt before they could be attacked.
The incident follows another Cosmos software disclosure earlier this year, when a researcher revealed a CometBFT flaw rated CVSS 7.1 that could stall nodes during block synchronization. That issue did not allow direct asset theft, but the two disclosures together have raised questions about the coordination between the organizations that maintain core infrastructure and the networks that depend on it.
## The governance dilemma
At the heart of the controversy is a genuinely hard tradeoff. Disclosing a critical vulnerability to dozens of network operators increases the chance that everyone patches quickly, but it also increases the chance that the details leak to attackers before the patch is deployed. Silent patching protects against exploitation in the short term but leaves operators flying blind when the patch lands, unable to prioritize what looks like a routine upgrade.
Cosmos Labs chose the second path, based on an assessment that production networks were not at risk. When that assessment proved wrong, the fallback was a race between operators updating quickly and attackers reverse engineering the obscured fix. The attackers won that race at least six times.
The episode also exposes the structural reality of appchain ecosystems. In a single-chain world, one team can coordinate a hard fork on its own schedule. In a world of dozens of sovereign networks sharing framework code, every critical patch becomes a coordination problem across independent validator sets, each with its own governance timelines, testing procedures, and risk tolerances.
## What comes next
Cosmos Labs’ advisory classified the flaw as critical and identified Cosmos EVM releases before v0.6.2 and v0.7.2 as vulnerable, giving remaining networks a clear upgrade target. The company has published a technical post-mortem and transaction details, which will inform both insurance claims and any potential recovery efforts.
The longer-term consequences will be institutional. Networks running shared infrastructure are already discussing security advisory processes that sit between full disclosure and silence, including embargoed operator notifications with delayed public disclosure. Whether the ecosystem adopts such a standard, or repeats the same gamble on the next critical bug, may determine whether cross-chain attacks like these remain a quarterly occurrence or become rarer.
For the affected networks, the immediate work is accounting. The 5.72 million USD in converted assets has moved through exchanges, and tracing continues. For everyone else building on shared frameworks, the lesson is blunt: a patch you were not told about is a vulnerability you still have.
Disclaimer: This article is for informational purposes only and does not constitute financial advice.
targeting burn addresses and multisigs because the funds cant flee. that is planning, not some kid scanning rpc endpoints
A silent patch left MANTRA, TAC and KiiChain blind for months. Who exactly decided quiet shipping was acceptable here?
the bug is textbook. delegate more than the balance holds, wrap below zero, get 2^256-1 base units for free
delegate more than you hold, wrap below zero, collect 2^256-1 units. it worked for months because those balances were assumed unmovable
months of blind networks and disclosure only after the draining starts. standards in this industry are a joke
cosmos labs shipped a fix in april, validated it privately, and let operators run exposed validators for months. that decision needs names attached to it
names attached, yes. april patch, august disclosure, that gap was a choice someone signed off on
april patch to august attacks is the part that should scare operators. the fix existed, attackers just found it faster than the disclosure did
attackers diffing april commits while operators scrolled past them. the info was technically public the whole time, just buried in a changelog nobody reads
it is, and the failure mode was the patch shipping tagged as routine so no security pipeline flagged it. disclosure norms dont help when the fix arrives disguised as chores
a patch existed since april and nobody told operators. silent patches are how you get 5.72M drained
patch shipped in april, disclosure only after 5.72M moved. someone at cosmos labs understood the blast radius and stayed quiet anyway
right?? cosmos labs couldnt reproduce it so they assumed funds were safe. four months of networks flying blind
four months blind because cosmos labs couldnt reproduce it. isnt that the exact scenario coordinated disclosure exists for
the whole point of coordinated disclosure is telling operators before attackers demo it for you. four months of silence is not a process, its luck
w take. disclosure norms in cosmos need a full rethink after this
the fix landed without a security advisory tag, so half the sync pipelines pulled it silently and half didnt. worst of both worlds
half synced, half didnt, and nobody could tell which was which. silent patches are how you get 5.72M drained across six chains in five days
same here, our sync pipeline pulled the april commit as a routine bump. nobody knew it was a security fix until the disclosure landed this week
2^256-1 wraparound on a delegated balance is wild. scarier part is they targeted burn addresses and multisigs that cant move funds
burn addresses and multisigs was the tell imo. whoever wrote this scouted which accounts literally couldnt move funds to fight back before pulling anything
5.72M across six chains from one integer underflow. every evm fork of evmos just got a free audit checklist item
honestly every evmos fork out there should be grepping for the same underflow pattern this week. if your chain synced the april patch late you are the low hanging fruit now
5.72M washed through dexes and cexes over five days and nothing got frozen mid-flight. that launder trail has to be a mess for whoever is tracing it
tagging the fix as routine is the part that gets me. one advisory line in april and MANTRA and TAC operators would have synced the same day