Agentic AI for Digital Asset Risk: What Finance Teams Should Ask
A major blockchain analytics provider has declared that risk decisions in digital asset finance will soon outrun human capacity, and that systems, not faster analysts, must make the routine ones. For finance leaders choosing crypto accounting software, the practical point is simple: if automated agents start making or supporting risk calls, every one of those calls needs an evidence trail you can defend to an auditor or a regulator.
What Elliptic announced and why it matters to finance teams
Elliptic, which sells blockchain analytics and screening tools to exchanges, banks and compliance teams, published a strategy piece the day after announcing a Series D funding round. The company says it is committing to what it calls an agentic operating model for digital asset risk. In plain terms, software agents would handle routine risk cases end to end, and people would be reserved for the ambiguous ones.
This is a vendor describing its own roadmap, so treat it as a market signal rather than independent evidence. The claims about market size, customer share and time savings come from the company itself, and the piece gives no third-party verification of them. What makes it worth reading is the direction of travel it describes, because that direction affects how firms should evaluate any tooling that touches transaction data, including accounting and bookkeeping platforms.
The core argument: volume has outgrown headcount
The company's reasoning runs like this. Stablecoin volumes reached $33 trillion in 2025, according to the piece, and banks and payment networks have moved from pilots to building on chain. Prediction markets and on-chain settlement are described as shifting from edge cases to default rails. Every flow generates risk decisions, such as whether to screen, flag, hold or release, and the company argues the rate of those decisions can no longer be absorbed by human teams.
It also argues that hiring is not the answer. The talent pool of genuine digital asset risk specialists, it says, has not kept pace with demand. Whether or not you accept that framing, anyone who has tried to staff a crypto compliance or crypto accounting function will recognise the pressure.
From faster analysts to systems that decide
Most vendors, the company says, pitch AI as a way to help an analyst work faster. It calls that frame too small. It cites a reduction in the time an entry-level analyst spends researching each alert from five minutes to under one, which it translates to roughly three hours a day saved across a team. But it argues the real shift is routine cases handled by agents, with escalation of the novel typology, the cross-jurisdictional pattern and the decision that must be defended to a regulator.
The evidence trail is the real compliance requirement
The most useful sentence in the piece for accountants is the one about defensibility. The company states that a risk decision in financial services has to be defensible, and that a risk score alone is not. Its stated approach is to return the reasoning, the evidence and the policy citations behind each decision.
That is the standard finance teams should hold every automated system to, whether it screens wallets or classifies transactions in a ledger.
What an auditor will want to see
If an automated agent contributes to a risk decision that touches your books, an auditor or examiner will reasonably ask several questions. They apply equally to a screening agent and to an accounting automation.
- What data did the system rely on, and how current was it?
- Which policy or rule did it apply, and who approved that policy?
- Can the decision be reproduced later from stored inputs?
- Who reviewed the outcome, and when was it escalated to a person?
- Where can a human override the result, and is the override logged?
The company says risk appetite, thresholds and accept or reject rules belong to the customer, and that it does not bake in defaults a customer cannot override. That is a sensible design principle. For your own governance, it means the accountability for thresholds stays with your firm, not the vendor, even when an agent applies them.
Data quality sits underneath everything
The piece makes a point that finance teams can reuse: a thin dataset behind a model simply makes it faster at being wrong. The company positions its curated attribution data as the foundation that makes automation safe, and says two-thirds of global cryptoasset volume flows through exchanges built on its technology. That figure is the company's own claim and I have not been able to verify it independently.
The principle holds regardless of vendor. Automated classification of wallet activity, counterparty risk or transaction purpose is only as reliable as the underlying labels. If you use crypto compliance reporting workflows that depend on counterparty attribution, you should know where those labels come from and how errors are corrected.
Implications for accounting firms, auditors and CFOs
For firms and CFOs buying crypto accounting software
The announcement does not describe an accounting product. It describes risk and AML infrastructure. But the operating model it proposes will not stay in the compliance department. As agents handle more routine decisions, the same expectations will spread to bookkeeping and reconciliation tools. When you evaluate crypto accounting software, or digital asset accounting software more broadly, treat agentic features as a governance question first and a productivity question second.
Useful diligence questions include the following.
- Does the tool record why it classified a transaction a certain way, not just the result?
- Can your team set and change the rules, and is every change versioned?
- What happens to ambiguous items: are they queued for a person or silently resolved?
- Can the output be exported in a form your auditor can test?
- Does the vendor expose data in a way that your own controls and agents can read?
On that last point, the company says customers will build their own agents and that the intelligence layer must be exposed in patterns agents can reason over, not just rendered as a screen for humans. For finance teams, that implies APIs and structured outputs will matter as much as the interface. It also raises access-control and change-management obligations, because an agent that can read and act on financial data needs the same segregation of duties you would apply to a staff member.
For bookkeepers and smaller practices
If you run a smaller practice and rely on crypto bookkeeping software, you are unlikely to deploy your own agents soon. The takeaway is narrower. Ask your software provider how automated decisions are documented, and keep your own record of the assumptions you accepted. If a tool auto-categorises activity and you sign off without review, the professional responsibility is still yours.
For individual filers
This announcement is aimed at institutions, not individual taxpayers, and the source says nothing about personal tax reporting. The indirect effect is worth knowing, though. As exchanges and payment providers automate risk screening, accounts can be flagged or restricted faster. Keep clean records of where funds came from and where they went, because an automated hold is easier to resolve when you can produce documentation quickly.
Machine-speed risk and the regulatory backdrop
The company closes its argument with several predictions. Autonomous agents, it says, will originate, authorise and settle a growing share of on-chain transactions, though not the majority by 2027. It also says institutions will not expose full balance sheets publicly, and that private rollups, confidential transactions and permissioned views will create zones where full data is unavailable. Risk infrastructure, it argues, must work across both transparent and privacy-shielded environments, and it says that work has begun.
The privacy gap is an accounting problem too
If transaction data becomes less visible on chain, the burden of evidencing it shifts to the entity holding the assets. Proof of ownership, provenance and valuation will rely more on internal records, attestations and permissioned views, and less on a public explorer anyone can check. Finance teams that currently lean on public chain data for existence and completeness testing should plan for the day that evidence is partly or wholly unavailable.
Regulators already expect governance of AI
The page the source came from points readers to material on the Financial Stability Board's AI sound practices and notes that regulators have made AI agent governance part of the compliance job. I have only the promotional pointer, not the underlying text, so I will not characterise the FSB's recommendations here. For our own coverage of that topic, see FSB AI sound practices and agentic risk in crypto. For a firm-level view of how automation fits into wider controls, our piece on enterprise risk management for crypto-active firms is a useful companion.
A practical checklist before you rely on agentic tooling
Steps to take this quarter
- Inventory every automated decision that touches crypto transactions, from screening to ledger classification.
- Assign a named owner for each rule set and threshold, and record approvals.
- Require evidence trails: inputs, rule applied, output, reviewer and override.
- Define which cases must always escalate to a person, and test that escalation works.
- Ask vendors in writing where their data comes from and how mistakes are corrected.
- Brief your auditor early so controls over automated decisions are tested rather than discovered late.
What to treat with caution
Vendor roadmaps are not regulatory requirements, and statements about market share or time savings are self-reported. Do not buy on promise alone. Pilot with a defined sample, compare automated outcomes against manual review, and document the error rate you observe yourself.
The broader point stands even if you discount the marketing. The volume of on-chain activity is rising, staffing is constrained, and automation is coming to every part of the finance stack. The firms that fare best will be those that can show, decision by decision, why a system did what it did.
Frequently asked questions
What does an agentic operating model mean in practice?
Per the company's description, software agents handle routine risk cases from start to finish and escalate ambiguous or novel ones to people. It differs from assistive AI, which only helps an analyst work through each case faster.
Is this an announcement about crypto accounting software?
No. It concerns digital asset risk and AML screening. It is relevant to accounting teams because the same expectations around automation, evidence and accountability will increasingly apply to bookkeeping and reconciliation tools.
Can I trust the figures in the announcement?
They are the vendor's own statements, including the stablecoin volume, customer share and analyst time savings. Treat them as claims, and verify anything material against independent sources before relying on it.
What should an auditor ask about automated risk decisions?
Ask what data was used, which policy was applied, whether the result can be reproduced, who reviewed it and whether overrides are logged. A score without reasoning and evidence is not enough to defend a decision.
Does this affect individual crypto filers?
Not directly. The source covers institutional risk infrastructure and says nothing about personal tax. Indirectly, faster automated screening can mean quicker account flags, so keep records of fund sources and transfers.
Source: Elliptic