When Cosmos Labs published its postmortem for the Cosmos EVM exploit wave on August 28, it put official numbers on an incident that had, until then, been measured chain by chain through the statements of its victims. The totals are more sobering than the public tally of named targets suggested — and the process failures behind them are more specific.
What the Official Account Confirms
According to the postmortem, attackers exploited the vulnerability between August 20, 19:06 UTC and August 25, 15:20 UTC — a five-day window that began with MANTRA and continued after the fix was public. Six networks were exploited, not the three that dominated headlines. MANTRA, TAC, and KiiChain were named by their own teams; the other three affected networks have not been publicly identified.
The realized losses: attackers exchanged stolen tokens for approximately $2.87 million in other assets on decentralized exchanges and sold an estimated $2.85 million on centralized exchanges, based on August 19 prices — roughly $5.7 million in total. That figure is far below the nominal value of tokens moved. The three publicly named chains alone reported token drains nominally worth around $21 million combined: 720.9 million MANTRA ($3.6 million), 2.99 billion TAC ($7.5 million per SlowMist's valuation), and 148.3 million KII (~$9.7 million). Illiquid tokens sold at heavy discounts explain most of the gap.
One enforcement-adjacent detail stands out: the centralized exchange accounts the attackers used have been frozen pending investigation by the relevant authorities, according to the postmortem.
The Triage Decision
The postmortem's most consequential admission concerns how the bug was handled before the attacks began. The vulnerability had been reported through Cosmos Labs' bug bounty program and was initially assessed as not presenting a risk of fund loss to production network configurations. On that assessment, it was addressed through the silent, public patch process — merged upstream on May 15 — rather than the private patch distribution used when a vulnerability is believed to threaten live funds.
The patched release, v0.6.2/v0.7.2, shipped one day before the first attack. Attackers watching the public repository could see exactly what changed, and exactly which production chains were still running vulnerable versions. KiiChain's own postmortem accuses Cosmos Labs of publishing the security fix before giving affected chains timely private notice — an interval in which multiple networks were hit.
The counterpoint comes from KiiChain's own outcome: its internal team spotted the anomalous drains and halted the network at 22:50:58 UTC on August 22, freezing 54.4% of the stolen 148.3 million KII on-chain. A halt initiated from inside the chain caught more of the damage than the warning issued from outside it.
Scope of Response and What Was Known
The mechanism itself — an unchecked underflow in the EVM StateDB's balance write-back for vesting accounts, wrapped to ~2^256 and weaponized through an overflow on a victim transfer — was covered in our earlier analysis of MANTRA's postmortem. What the official account adds is scale: Cosmos Labs coordinated with forty Cosmos chains to assess exposure, and thirteen other potentially exposed networks patched, halted, or applied other mitigations without further incidents.
TAC illustrates the cost of being downstream of a public fix without applying it. Its exploit began at 19:46:37 UTC on August 22 and the chain halted at 23:58:11 UTC — four hours and twelve minutes later, by which point the bulk of the drained funds had crossed a bridge to BNB Chain in roughly 95 seconds, per independent forensic analysis by Rarma.
The Audit Gap
The postmortem notes the vulnerable code path had been reviewed internally and by external independent firms, and that the exploited vulnerability was not discovered in those audits. Since January 2025, Cosmos Labs has triaged thousands of vulnerability reports and paid more than $850,000 to researchers through its bounty program. The failure was not a missing review — it was a misjudged severity classification that sent a fund-draining report down the wrong disclosure lane.
The commitments are procedural: better identification of reports whose scope exceeds what the original document describes, health checks on developer responsiveness in coordinated-vulnerability-disclosure channels, public standards for when those channels are used, and a long-term external audit of the ecosystem's security practices. Whether the next shared-dependency bug is handled privately and quickly will be the test of whether any of that took.
TrustGrade tracks the security posture of DeFi protocols and chain infrastructure. Security scans with verified, registry-backed scores arrive with TrustGrade Code Scoring in December 2026.