
$680 Million Warning: Most DeFi Attacks Fall Outside the Scope of Audits
By Liam 'Akiba' Wright
Translated by Chopper, Foresight News
In decentralized finance, "audited" is often treated as a security endorsement for an entire project. In reality, audits typically only cover specific points in time, designated code, components, and versions. Anything added, removed, or modified outside this scope can render the audit results entirely different.
A new preprint puts concrete numbers on this gap. Researchers from security firm ack3 and the Czech Technical University in Prague analyzed 135 security incidents reported in H1 2026, which resulted in losses totaling $939.86 million. They found that 68 of these incidents had identifiable public pre-incident audit records.
Of these 68 incident samples, researchers classified 46 attack vectors as falling completely outside any verifiable audit scope; 20 incidents fell within at least one audit coverage area; the remaining 2 were undetermined. Incidents outside audit scope accounted for 67.6% of the total incident count, yet corresponded to 94.4% of the reported loss amount.
This staggering ratio is not an evaluation of audit effectiveness, nor evidence that scope limitations caused the losses. It merely represents the loss distribution within the selected sample of public security incidents. Two major events heavily skewed the data: excluding the $292 million loss from Kelp DAO and the $285 million loss from Drift Protocol, the share of losses from out-of-scope attacks within the same audited sample dropped to 72.1%.
Despite these limitations, the study highlights a foundational security trust issue: a project may claim to be audited, but users cannot know whether the actual deployed systems, fund flows, and related control mechanisms have been reviewed.
The True Meaning of the Data
ack3’s research covers January 1 to June 29, 2026, comprising 122 confirmed attack incidents and 13 suspected ones. Across the full dataset, 35 incidents lacked audit records, and 32 had unknown audit histories; both categories are excluded from the aforementioned 68-incident sample.
Within the 68-incident sample, out-of-scope events accounted for $680.97 million in losses against a total of $721.24 million, yielding the 94.4% figure. Excluding Kelp DAO and Drift Protocol, out-of-scope losses amounted to $103.97 million against a total of $144.24 million, representing 72.1%. The dataset json file allows for reproduction of the incident classification counts and loss amounts.
The "in-scope/out-of-scope" labels are determinations made by researchers based on public evidence. The team reviewed project and auditor archives to locate post-attack audit reports, comparing the final attack vectors against reviewed code, versions, and auditor exclusions. This research is a six-page preprint co-produced with the dataset publisher, authored by two individuals affiliated with security audit firm ack3.
The study lacks an unattacked control group and does not track how long each system was exposed to risk. Consequently, it cannot prove that audited protocols are inherently safer overall, cannot estimate incident probability, and cannot confirm that "falling outside audit scope" was the direct cause of every loss. Some unpublicized audits and private incidents may be missing, and reported loss data is not fully comparable.
Therefore, the study can only draw limited conclusions: audit records and audit coverage are two separate metrics. A reviewed smart contract does not guarantee equivalent security for contract upgrades, privileged keys, frontends, relays, oracles, cloud services, or incident response procedures.
Two incidents that occurred in August corroborate this distinction from different angles. The ICON Network case clearly demonstrates how two pieces of reviewed code failed at the boundary between two verification processes; meanwhile, the August aelf security incident presents a different scenario where current audit evidence cannot yet map the attack execution path to a pre-incident audit scope.
ICON Network: A Failure Sample at Audit Boundaries
During the August 27 ICON Network replay attack, two modules in the withdrawal pipeline interpreted the same message differently.
According to ICON Foundation’s post-incident report: the migration contract relies on the high-order bits of the withdrawal message sequence number to determine message uniqueness; however, the cryptographic signature only covers the lower 256 bits of the sequence number. By modifying the high-order bits excluded from signature verification, an attacker resubmitted two legally signed withdrawal messages 1,492 times over approximately 20 minutes, with 1,490 executions succeeding.
The replay attack drained 119.866 million ICX and 531,600 bnUSD. At the time of the report release, ICON confirmed a net loss of approximately 150.2 ETH plus 31,204 USDC. The Foundation stated that 531,600 bnUSD and 1.366 million SODA assets have been recovered, and user deposits, account balances, and positions remain unaffected.
ICON stated that this migration contract underwent external auditing, implementing audit recommendations including relevant modifications in the same module area; the corresponding relay logic also received dedicated independent review. The Sodax developer documentation audit list includes eight reports covering various components, including a November 2025 Sodax relay audit report.
However, the incident report explicitly notes that this precise mismatch between the uniqueness verification logic and the signature verification value falls outside the scope of the aforementioned audit findings. A simple "Audited" project tag cannot inform users whether both ends of the withdrawal pipeline maintain consistent standards for "message uniqueness."
The response timeline also exposes another type of boundary issue. ICON's first automated alert triggered at 02:08 UTC, roughly 7 minutes after the attack began. Staff initiated investigations around 03:40, suspended the affected contracts at 03:53, and paused the entire network at 06:18:54.
Approximately 90 minutes passed between the initial alert and full incident response. ICON attributed this to adjustments in its alerting mechanism, as the rule generated numerous false positives during past network connectivity issues and thus did not trigger high-priority notifications for on-call personnel. The Foundation plans to deploy automatic shutdown triggers, lower circuit breaker thresholds, and conduct specialized reviews focused on message uniqueness and replay protection.
Such risk control mechanisms cannot replace audits, but they address another critical question: when preventive measures fail, can the system rapidly detect and isolate risks?

Users still need clear answers to fundamental security questions
aelf: Security Guarantees Require Continuous Updates
The August security incident involving aelf corroborates the aforementioned point from another angle. Public records describe a runtime intrusion followed by controlled recovery, but existing evidence is insufficient to determine whether the attack path fell within a specific pre-incident audit scope.
An official project announcement states: an unauthorized smart contract exists capable of injecting encoded .NET assemblies and commands into the node execution pipeline via transaction parameters.
The preliminary investigation report attributes the incident to flaws in runtime reflection and dynamic loading verification, alongside insufficient isolation between the contract execution environment and sensitive node/infrastructure resources. aelf identified a total of 155 related transactions across five independent payload assemblies, which possessed capabilities to execute host commands, attempt external communications, access node keys, and perform infrastructure reconnaissance.
Possessing these capabilities does not confirm that all payloads executed successfully, nor does it mean attackers obtained all target credentials or that sensitive data was leaked. aelf stated that signature keys and infrastructure credentials have been rotated according to potential exposure protocols.
As of September 11, this assessment remains provisional. The aelf official blog has published no specific updates regarding this incident since August 26. The August 26 announcement promised subsequent updates and a final post-mortem.
aelf’s technical security documentation notes that its blockchain and ELF token contracts have undergone multiple rounds of auditing with no security issues found. However, current public pages cannot map the August attack's runtime path to any pre-incident audit report. Consequently, classifying the incident as either an audit oversight or an out-of-scope failure lacks sufficient evidentiary support.
This uncertainty itself holds reference value. A timestamped audit report gradually becomes misaligned with current system code, dependency libraries, and actual operational states. Users require version-controlled security records to reflect these discrepancies.
This security record should specify: the audited repository and code commit versions, deployed contract addresses, excluded components, privileged roles, and dependencies; while simultaneously recording post-audit contract upgrades, key custody and rotation mechanisms, runtime isolation strategies, alerting and circuit breaker mechanisms, and timestamped asset recovery statuses, distinguishing confirmed losses, frozen assets, and unresolved risk exposures.
This is not to diminish the value of audits, but rather to align audit marketing with actual work content and tie it to currently operating systems.
An audit badge cannot answer whether audited components, deployed systems, and failure response mechanisms remain within the same security perimeter.













