Elliptic Launches Decode: AI Agent for Blockchain Intelligence
Elliptic launched Decode on 17 September 2026, an AI agent that accepts plain-language questions about any blockchain address, entity, or transaction and returns a sourced, ranked answer from the firm's intelligence platform. For compliance teams at financial institutions and accounting firms running digital asset books, the launch represents a meaningful shift in how on-chain risk data can be accessed and, critically, how that data might feed into crypto accounting software and audit workflows.
What Decode Actually Does
The core proposition is straightforward: instead of navigating technical APIs or proprietary dashboards, a user types a question in plain English. "Who does this address belong to?" or "Which counterparties has this exchange dealt with?" are the kinds of queries Elliptic cites. The agent then decides, autonomously, whether to query a dataset directly by generating and executing SQL, or to call the API endpoint best suited to answer that specific question.
Coverage and data depth
Decode draws on Elliptic's intelligence platform, which the firm says covers more than 68 blockchains and holds billions of attribution records. That breadth matters for multi-chain accounting environments where a single client portfolio may span Bitcoin, Ethereum, Solana, and several Layer 2 networks simultaneously. The agent returns results as a ranked, filterable table when the answer involves multiple records, and as prose when the finding is qualitative or concerns a single entity.
How the agent reasons
Users can observe every reasoning step in real time as the agent works through a query. This transparency is deliberate. Elliptic frames it as "human oversight by design," a principle it has codified in what it calls the Elliptic Standard. If a question cannot be settled by the available data, the agent declines to answer rather than speculate. The firm notes that a small number of queries will be refused at this stage, on the basis that a refusal is more useful than a confident but unsupported response. Context is retained across a session, so a compliance analyst can layer follow-up questions without re-entering background information.
Rollout Timeline and Access
Government customers have access to Decode immediately. Crypto compliance teams at financial institutions are in a second wave, with access planned before the end of 2026. The third phase opens the agent to external software integrations, meaning systems can submit queries programmatically rather than requiring a human to type them. After that, Elliptic plans to embed Decode inside its existing Lens and Investigator products.
What the phased rollout signals
The sequencing reflects the sensitivity of the underlying data. Restricting early access to government users and regulated financial institutions gives Elliptic a controlled environment to stress-test the agent's refusal logic and output quality before wider exposure. The software-integration phase is arguably the most consequential for the accounting and audit sector, because it opens a direct pipeline between Elliptic's intelligence layer and third-party crypto accounting software platforms that firms already use for bookkeeping and transaction classification.
AML and Compliance Implications
AML teams today typically face a two-step problem: they need to identify whether a counterparty or wallet address carries risk, and they need to document that assessment in a form auditors can rely on. Decode compresses the first step by surfacing attributed risk data without requiring a specialist to construct a manual API call or database query. The agent's visible reasoning chain also provides a natural audit trail, the thinking steps it exposes in real time can be captured and attached to a suspicious activity report or a know-your-customer file.
Where human judgment still applies
Elliptic is explicit that Decode does not close cases and does not determine investigation outcomes. It surfaces what the data supports; a human analyst makes the consequential call. This is not just a liability hedge. Financial Action Task Force guidance and most national AML frameworks require that automated screening tools remain subject to meaningful human review. The agent's design is consistent with that expectation, it produces evidence, not decisions.
For compliance officers at banks and virtual asset service providers, that distinction matters when regulators examine whether a firm's transaction monitoring system constitutes a genuine control or a rubber stamp. The ability to show an examiner the full reasoning chain behind a risk assessment, including what data sources the agent queried and why, strengthens the credibility of the control environment. This is also directly relevant to the challenge explored in our earlier analysis of how blockchain spam transfers complicate sanctions screening, where the quality of attribution data is precisely what separates a defensible alert from a false positive.
Accounting and Audit Firm Considerations
For accounting practices with digital asset clients, Decode's eventual software integration phase is the development worth tracking most closely. Once the agent is accessible via API, it becomes possible to build automated enrichment into a transaction ingestion workflow: an address is imported into the firm's crypto bookkeeping software, a query fires to the Decode agent, and a risk classification is returned and logged before a human reviewer ever opens the file.
Audit evidence quality
International Standard on Auditing 500 requires auditors to evaluate the reliability of the sources from which audit evidence is obtained. AI-generated outputs from a third-party intelligence platform will need to be assessed against that standard. The relevant questions are whether the underlying dataset is independently verifiable, whether the agent's reasoning is reproducible, and whether the output is sufficiently specific to support a conclusion about a particular transaction or counterparty. Elliptic's decision to expose reasoning steps addresses the reproducibility point directly, but firms will still need to form a view on dataset provenance and coverage.
CFO and treasury implications
CFOs overseeing treasury positions in digital assets have a different but related concern. Counterparty risk assessment prior to a large on-chain settlement has historically required either an in-house blockchain analytics capability or a manual request to an intelligence vendor. An agent that can answer a counterparty question in seconds, with a documented evidence chain, changes the economics of that process. It also raises the bar for what constitutes adequate pre-trade due diligence, because the capability now exists to perform it at speed and at scale.
This dynamic is not isolated to Elliptic's product. The broader pattern of AI agents being layered onto blockchain intelligence infrastructure is one that compliance-focused firms need to track as part of their digital asset accounting software evaluation cycles. For comparison, the expansion of on-chain coverage by other intelligence providers follows a similar logic, as covered in our piece on Chainalysis Arc blockchain AML coverage.
What Firms Should Do Now
The practical steps differ by firm type, but the underlying priority is the same: understand where Decode fits in an existing compliance or accounting workflow before the software integration phase arrives.
For compliance and AML teams
If your institution is an existing Elliptic customer, the firm directs current users to contact their customer success representative. Early engagement allows compliance teams to shape how the agent is configured for their specific risk appetite and to begin building the internal procedures that will govern how Decode outputs are reviewed, documented, and stored. Waiting until the software integration phase to think through those procedures is leaving it late.
For accounting firms and auditors
Firms advising digital asset clients or conducting audits of entities holding crypto assets should begin assessing how AI-generated blockchain intelligence outputs will be treated within their evidence frameworks. That means drafting a position on what corroboration, if any, is required alongside a Decode output before it can be relied upon as audit evidence. It also means updating engagement letters and client onboarding questionnaires to capture whether clients are using Elliptic's products, because the outputs may surface in client-prepared documentation during an audit.
For CFOs and finance teams
Teams managing digital asset treasuries should evaluate whether programmatic access to on-chain intelligence, once available, should be integrated into their pre-settlement approval workflow. The case for doing so is strongest where settlement volumes are high and counterparty diversity is broad. Document the decision either way, because regulators in multiple jurisdictions are increasingly examining whether treasury processes for digital assets are proportionate to the risks involved.
Frequently Asked Questions
What is the Elliptic Decode agent?
Decode is an AI agent launched by Elliptic that allows users to ask plain-language questions about blockchain addresses, entities, and transactions. It queries Elliptic's intelligence platform autonomously, either by generating SQL against a dataset or by calling the appropriate API endpoint, and returns a sourced, evidenced answer.
Who can access Decode right now?
As of the launch date of 17 September 2026, access is available to government customers. Crypto compliance teams at financial institutions are next in line, with access expected before the end of 2026. Software integrations for programmatic access come after that.
Does the agent make compliance decisions?
No. Elliptic has been explicit that Decode surfaces intelligence and leaves judgment with the human analyst. It will not answer questions the data cannot settle and will not provide opinions outside its evidential scope. This is consistent with FATF expectations that automated monitoring tools remain subject to meaningful human review.
How does Decode affect crypto accounting software workflows?
Once the software integration phase opens, compliance and accounting teams will be able to build programmatic queries into their transaction ingestion pipelines. An address imported into a digital asset accounting software system could automatically trigger a Decode query, returning a risk classification that is logged before human review begins. This compresses the manual enrichment step that currently sits between transaction import and risk assessment.
What should auditors consider when Decode outputs appear in client files?
Auditors should evaluate Decode outputs against ISA 500 criteria for audit evidence reliability. Key considerations include the verifiability of Elliptic's underlying dataset, the reproducibility of the agent's reasoning (which the visible thinking-step log supports), and whether the output is specific enough to support a conclusion about a particular transaction or counterparty. Firms should establish a written position on this before such outputs start appearing in client-prepared documentation.
Source: Elliptic