Trace Desk: agent fund tracing where every hop comes with onchain proof
Agents trace stolen or laundered crypto and attach a verifiable proof to each hop. Exchange compliance teams pay.
Crowded · ABF-Bench makes tracing verifiable
Written for: Small startup team
Next step
Call an incident-response lead at a crypto security firm that handled a Q3 hack, and ask how many analyst-hours one trace took and which steps they would hand to an agent if every hop came with an onchain proof.
Why now
Q3 hacks cost $1.18B, and Bank of America's $1.2B fine for unreported suspicious activity shows that enforcement on monitoring failures is getting heavier. Exchanges handling freeze requests need faster traces. But the benchmark is new, agent accuracy is unproven, and Chainalysis and TRM own the category, so this is a watch.
What became possible
ABF-Bench is an interactive benchmark for agentic blockchain forensics. Because public ledgers record every movement, each claim an agent makes while tracing funds through splits, swaps and bridges can be independently verified.
One way in
- Trace Desk
Agent that follows funds from a hack or scam address. It produces an evidence chain where each hop links to the transaction proving it, and stops for a human at mixers and custodial services.
- Customer
- Compliance teams at mid-size exchanges and crypto incident-response firms
- Model
- Per-case fee, plus a subscription for in-house teams
Competition
Both openings face real risks. The big vendors could add verifiable proofs quickly. The bigger risk is the data: their label data and attribution cover more of what investigators actually need than proof of individual hops does. A newcomer would probably have to lead with the evidence format, and with how well it holds up in legal and freeze-request settings, rather than with tracing itself.
Full competition notes (6 competitors)
This space is crowded. All three major blockchain intelligence vendors shipped tracing agents in 2026: TRM's Co-Case Agent in March, Chainalysis's agents rolling out from summer, and Elliptic's Decode in September. Each is pitched on the same points this idea uses: audit trails, evidence attached to each answer, and a human kept in control. AMLBot and MetaSleuth cover the cheap, self-serve end. Linking every hop to the transaction that proves it, and pausing for a human at mixers and custodial services, is now standard, so neither can be the main selling point. One opening that may still exist is cryptographic proof that a third party can check without trusting the vendor. That could mean light-client, Merkle or receipt inclusion proofs for each hop, packaged as a portable evidence bundle. An exchange receiving a freeze request, a court or a counterparty could then verify the bundle independently instead of relying on TRM's or Chainalysis's database. The bundle would also need to keep the deterministic tracing facts (provable) visibly separate from heuristic attribution (labels and clustering, which can't be proven). I found nobody explicitly selling that. A second opening is price. Mid-size exchanges that can't justify TRM or Chainalysis seat costs could pay for an exportable, standards-based evidence format that works with existing tools, for example as a layer that ingests their exports. Both openings face real risks. The big vendors could add verifiable proofs quickly. The bigger risk is the data: their label data and attribution cover more of what investigators actually need than proof of individual hops does. A newcomer would probably have to lead with the evidence format, and with how well it holds up in legal and freeze-request settings, rather than with tracing itself.
- TRM Labs Co-Case Agent
Launched March 2026 inside TRM Forensics. It turns natural-language prompts into tracing actions, suggests next steps, and keeps an immutable audit log meant for SAR filings and court use. TRM says the agent does not change graphs on its own, so the investigator stays in control. It comes at no extra cost to existing customers, who include exchanges. TRM recently raised a $70M Series C to build orchestrated AI agents.
- Chainalysis blockchain intelligence agents
Announced at Links in March 2026, with rollout starting summer 2026 for investigations and compliance first. The agents handle multi-chain tracing, alert enrichment and intelligence reports, built on more than 10 million past Reactor investigations. Chainalysis describes them as deterministic, auditable and grounded in verified onchain data, with a human kept in the loop.
- Elliptic Decode agent and Investigator
Decode launched in September 2026. It answers onchain risk questions and attaches the evidence behind each answer. It is live with government customers now, and Elliptic says the private sector gets it later in 2026. Elliptic Investigator already generates fund-flow graphs automatically and traces across chains and bridges.
- AMLBot AI Tracer
Launched July 2026. It is a self-serve tool: you submit a transaction hash and it follows funds through intermediate wallets, splits and bridges, then labels where they end up. It is aimed at the low end of the market, including victims and smaller firms.
- BlockSec MetaSleuth
Cross-chain fund-tracing tool covering 12 networks, with more than 600M address labels and monitoring of funds that have not moved yet. It is widely used after hacks to follow stolen funds to exchanges. I found no confirmed AI agent features.
- Bitquery Crypto Investigation Services
Investigation service run by people. It traces hacked funds through mixers and bridges to exchanges and attributes the addresses involved. It competes for the incident-response budget, not as an automated product.
Checked 2026-10-19
Risks
- Crowded: Chainalysis, TRM and Elliptic hold the data and the relationships
- Agent accuracy on adversarial laundering is unproven
- Address-attribution data, not tracing logic, is the real moat
Best on market pull and team fit; open space is the weak point (3).
tech 4 · pull 6 · space 3 · fit 6 · build 5
Exposed
Companies that could benefit or be threatened. News and context, not investment advice.
Momentum of its main topic
Sources
Status history
- 2026-10-19 Added to the watchlist