ABF-Bench tests agents on verifiable blockchain fund tracing
Original title: ABF-Bench: A Verifiable Benchmark for Agentic Blockchain Forensics
A memory harness reached 75% edge recall, versus 47% for Codex and 10% for Claude Code.
ABF-Bench turns evidence-backed fund tracing into an interactive agent task built from real laundering activity. Tasks cover multi-hop transfers, DeFi exchanges, cross-chain trips, custodial and mixer deposits, and address poisoning. A proposed harness with working memory beat general-purpose agents and recognized undisclosed custodial services that fixed-order tracers missed, but agents stopped early and trailed the best tracer on larger tasks.
- The harness hit 75% edge recall and 98% fidelity.
- Claude Code cited unretrieved records and fell to 10% recall.
- Custodial-service recognition was 89% for the harness versus 0%.
- Agents acquired evidence faster early, then lost recall on larger tasks.
- Tracing stops where ledger records leave a continuation unresolved.
From the original article
1 min read at arxiv.orgBlockchain forensics is adversarial and open-ended: money launderers split, swap, and bridge funds, cut their trail at mixers or custodial services, and change tactics from one incident to the next. Investigators must keep adapting their tracing in response. Yet the immutability of public ledgers preserves every recorded movement in public view, making much of the investigation independently verifiable.
By Xiaohang Yu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.