Audited DeFi projects lost $885M to out-of-scope attacks

Researchers affiliated with security firm ack3 and the Czech Technical University in Prague analyzed 135 reported DeFi incidents from Jan. 1 through June 29, 2026, attributing $939.86 million in total reported losses. The team identified public pre-incident audits for 68 incidents and mapped attack paths against audited code, components and versions.

Within that 68-incident subset the researchers classified 46 attack paths as outside every audit scope they could locate, 20 as inside at least one scope, and two as unresolved. The outside-scope cases made up 67.6% of the audited incidents and accounted for $680.97 million of $721.24 million in reported losses in that group, a 94.4% share. Two large incidents — Kelp DAO at $292 million and Drift Protocol at $285 million — account for a large portion of the dollar totals; excluding those two lowers the outside-scope share to 72.1% of losses in the audited set.

The ack3 dataset labels 122 incidents as confirmed and 13 as likely. Of the full set of 135 incidents, 35 had no identified audit and 32 had unknown audit histories; those cases were not part of the 68-incident comparison. To reach their classifications, the researchers searched project and auditor archives for pre-incident reports, then compared the reported attack paths with reviewed repositories, commits, deployed addresses and stated exclusions.

The paper is a six-page preprint produced with the dataset publisher and lists two authors affiliated with ack3. The researchers note the study does not include a comparison group of audited systems that were not exploited and does not measure exposure time for each protocol. The analysis does not claim audits cause fewer incidents, and the authors acknowledge that undisclosed audits and private incidents may be missing and that reported loss figures are not strictly comparable across cases.

Two August incidents are detailed in the report. In ICON Network’s Aug. 27 replay exploit, a migration contract used the high bits of a withdrawal message’s serial number to decide uniqueness while the cryptographic signature covered only the low 256 bits. By changing the unsigned high bits, an attacker resubmitted legitimately signed withdrawal messages 1,492 times over about 20 minutes; 1,490 calls succeeded, releasing 119.866 million ICX and 531,600 bnUSD. ICON’s postmortem put confirmed net loss at about 150.2 ETH plus 31,204 USDC and reported recovery of 531,600 bnUSD and 1.366 million SODA. The postmortem also shows the migration contract and related relay logic had received external reviews but that the precise mismatch between uniqueness checks and the signed value fell outside those findings. The incident timeline shows the first automated alert at 02:08 UTC, a staff investigation at 03:40, a pause of the affected contract at 03:53 and a full network halt at 06:18:54; the project plans lower circuit-breaker thresholds and automatic shutdown triggers.

Aelf’s Aug. 18 incident involved a reported runtime compromise in which an unauthorized contract could deliver encoded .NET assemblies into node execution paths. The incident report links the activity to gaps in checks for runtime reflection and dynamic loading and to weak isolation between contract execution and sensitive node resources. Aelf reported 155 associated transactions and multiple payload assemblies with potential host-level capabilities, and the team reported rotating signing keys and infrastructure credentials. Public audit records show prior reviews of the blockchain and token contracts, but available reports do not tie a specific pre-incident audit to the runtime path described in August.

The researchers recommend that public assurance records be versioned and specific: naming reviewed repositories and commits, deployed addresses, excluded components and privileged roles; recording dependency versions and upgrades since review; documenting key custody and rotation policies; and listing alerting and circuit-breaker behavior and dated recovery status so users can compare reviewed artifacts with deployed systems and incident-response arrangements.