"Audited" is often presented as a verdict on an entire project. In practice, a security audit covers named code, components and versions at a specific point in time — and everything added, excluded or operated around that boundary may carry a very different level of assurance. A recent preprint now puts hard numbers on that gap, and they are uncomfortable.
The study, "The ack3 H1 2026 DeFi Incident Dataset: Audit Scope Across 135 Security Incidents", was posted to arXiv on August 13 by researchers affiliated with cybersecurity company ack3 and the Czech Technical University in Prague. It cross-references public audit archives against the attack paths of incidents reported between January 1 and June 29, 2026.
What the Data Shows
The dataset covers 135 reported DeFi security incidents with $939.86 million in attributed losses — 122 graded confirmed and 13 likely, according to an analysis by CryptoSlate. The researchers could identify public pre-incident audits for 68 of those incidents.
Within that audited subset, the results split three ways: 46 attack paths fell outside every audit scope the researchers could identify, 20 fell inside at least one scope, and 2 could not be resolved. The outside-scope group represented 67.6% of incidents by count — but 94.4% of reported losses.
Per the CryptoSlate reading of the machine-readable ledger, outside-scope incidents accounted for $680.97 million of $721.24 million in losses across the 68-incident subset.
The Two-Outlier Caveat
The loss-weighted figure is dominated by two large incidents — identified in the dataset as $292 million at Kelp DAO and $285 million at Drift Protocol. Excluding both, the outside-scope share of losses falls to 72.1% ($103.97 million of $144.24 million). The direction of the result survives; the headline number does not.
That caveat matters, and the authors state it plainly. The study has other limits worth noting: the inside-or-outside labels are the researchers' judgments based on public evidence; there is no unexploited comparison group, so the work cannot establish whether audited protocols are safer overall or estimate incident probability; undisclosed audits and private incidents may be missing from the corpus. Two of the authors are affiliated with ack3, which sells security reviews — a conflict the paper's provenance makes visible, and a reason the dataset itself was published openly with a Zenodo DOI and a ledger searchable by victim, auditor, audit date and scope outcome.
Audit History Is Not Audit Scope
The study's central conclusion is narrow but consequential: project-level audit history and incident-path scope are distinct variables. A reviewed smart contract does not automatically confer the same assurance on an upgrade, a privileged key, a front end, a relayer, an oracle, a cloud dependency or an incident-response process.
For users, the practical takeaway is that an "audited" badge — however many firms and reports sit behind it — answers a different question than the one that matters: was the specific path holding my funds reviewed, and is that review still current with what is deployed?
This is the assurance-boundary problem that third-party scoring exists to address. TrustGrade tracks audit and security posture for protocols and platforms, and its registry-backed Code Scoring product — automated security scans paired with verified registry scores — is scheduled for December 2026. Details are at trustgrade.ai.