Nasdaq embeds AI agents in financial crime and capital markets workflows

Nasdaq’s AI relevance extends well beyond its exchange brand. Its financial technology businesses place AI inside financial-crime investigations, treasury and capital-markets operations. Verafin’s agentic analysts and Calypso’s governed AI environment illustrate two distinct approaches: specialized workers for a bounded investigation task, and an assistant operating against a platform’s system-of-record data. Both require clear evidence and an accountable person at the point where analysis becomes action.

01 / ProductA financial technology portfolio with distinct AI products

Nasdaq Financial Technology covers infrastructure used by financial institutions, including financial-crime and capital-markets operations. Nasdaq Verafin combines analytics with investigation workflows. It is part of the Nasdaq company story, rather than a separate new company in this directory. The relevant AI mechanism is the use of specialized agents to gather evidence, evaluate a defined alert and prepare work that staff can inspect.

The Verafin workforce page lists sanctions, anti-money-laundering, enhanced-due-diligence and fraud analysts. Its scope matters: cash-structuring is the stated current AML skill, and ACH is the stated current fraud skill, with additional capabilities planned. The June expansion announcement described a rollout across the second half of 2026. The current product page gives a more specific basis for evaluating a particular investigation than assuming all announced skills are already available.

Calypso addresses a different setting: front-to-back capital-markets and treasury operations. Its September 2026 launch introduced a governed agent environment and a natural-language assistant for platform data and documentation. A wider suite of operational workers is planned. Treat the available assistant, the integration framework and future automation as separate elements when assessing what a client can actually deploy.

02 / AudienceFor teams with established investigation and review processes

Verafin is relevant to financial institutions with defined alert queues, case records and experienced reviewers. A good first question is which repetitive research step consumes effort without requiring a new policy decision. The institution needs to know the evidence required to support its own disposition before asking an agent to assemble it. Otherwise the pilot risks automating an inconsistent process more quickly.

The ComplyAdvantage blueprint provides adjacent context for screening and risk information, while Quantexa explores entity context and connected data. These are complementary comparison points rather than identical agent products. A screening signal, a linked entity and a completed investigation are different outputs. Evaluate where each fits in the existing case process and what evidence transfers between them.

Calypso’s audience includes capital-markets operations and treasury teams already working with its records. An assistant that helps locate a trade record or interpret platform documentation has a different control profile from an agent changing a settlement instruction. Begin with the actual available capability and the institution’s approved access path. The existence of an extensible environment does not make every imagined trading or post-trade action an implemented feature.

03 / WorkflowA proposed sanctions-alert review workflow

Consider a proposed Verafin evaluation using a controlled set of previously reviewed sanctions alerts. The objective is to assess evidence preparation and reviewer usability, not to transfer responsibility for a legal determination. Sequenced has not tested this workflow. Select cases that include straightforward name mismatches, incomplete identifying information and genuinely ambiguous records, then establish an adjudicated reference set through the institution’s normal process.

Define exactly which records the agent may retrieve and what the reviewer must see. The case should identify the alert, relevant customer or counterparty fields, matched record and supporting source information. A narrative that omits the decisive date or identifier can look complete while being impossible to verify. Evaluate evidence coverage before considering whether the proposed disposition agrees with the previous reviewer.

Run the agent in a review-only configuration for the initial assessment. Keep automatic disposition disabled unless and until the institution has separately approved a tightly defined use case. Verafin describes both review and automation options, but availability is not a reason to select the most autonomous setting. Staff need a practical way to correct the rationale and retain the original output for comparison.

Examine disagreements case by case. Separate an incorrect match assessment from a missing input, stale source or policy interpretation. A discrepancy may reveal a flaw in the historical reference decision as well as in the agent. Escalate uncertain cases through the existing review path and record the resolution. The aim is a better understood case process, not a favorable agreement percentage assembled from easy examples.

Test operational exceptions as deliberately as substantive decisions. Include an unavailable data source, an interrupted request and an alert reopened after new information arrives. Confirm whether the case records show what the agent knew at the time. A later data refresh should not silently rewrite the evidence behind an earlier decision, because reviewers may need to explain why the original assessment was reasonable or incomplete.

Measure source completeness, correction effort, reviewer agreement on rationale and time to resolve ambiguous cases. For future AML or fraud pilots, create new task-specific sets rather than reusing the sanctions result. Cash structuring and ACH fraud involve different evidence and timing. Success on one skill cannot establish that another agent or a planned capability meets the institution’s requirements.

04 / PricingCommercial evaluation starts with a named capability

The Verafin workforce page offers an expression-of-interest route rather than public numeric pricing. Calypso similarly leads to a commercial conversation. The Calypso launch announcement says Nasdaq will work with individual clients through governance and control requirements. A product announcement therefore does not establish immediate activation for every deployment, especially when cloud and on-premises connections differ.

Ask whether licensing depends on institution size, case volume, selected agents or underlying platform modules. Separate integration, data access, support and reviewer training in the proposal. For an overlay on a third-party workflow, request a supported architecture and a demonstration of the case handoff. The existence of an overlay option does not establish compatibility with every case-management system or historical data format.

05 / DistinctionsSpecialization makes the evaluation more concrete

A named analyst with a bounded skill is easier to evaluate than an assistant promised to handle all financial operations. The institution can specify the required inputs, accepted evidence and escalation conditions. That specificity also creates a useful limit: a worker should recognize when a case lies outside its supported scope. An explicit escalation may be more valuable than a plausible answer to an unfamiliar problem.

Calypso offers a different advantage by locating assistance near a system of record. A query can be tied to governed trading, risk or collateral information rather than a detached document upload. The useful question is whether the response preserves the record’s identity and as-of context. The wider agent framework may support future applications, but the current assistant can be assessed on retrieval and explanation without assuming autonomous trade execution.

06 / QuestionsEvidence, permissions and escalation remain decisive

Ask which sources the selected analyst uses, how source freshness is communicated and which parts of its rationale can be reproduced later. Clarify who can change disposition policies and how overrides appear in the audit record. Request a demonstration of a case the agent cannot resolve. A polished success case says less about operational readiness than clear handling of incomplete or contradictory evidence.

For Calypso, confirm the exact hosting and connection arrangement, permitted data scope and available tools. The launch describes cloud-delivered agents and an intended connection path for on-premises clients, with client-specific governance work. Do not infer identical deployment readiness across those settings. Ask how a user’s platform permissions constrain an assistant response and whether future action tools will require separate approval.

07 / DecisionChoose a narrow task with an accountable reviewer

The next step is a demonstration built around a difficult, institution-approved case or query. Follow the source records, challenge the explanation and inspect the escalation path. Nasdaq is a substantial AI-related company because it connects these capabilities to established financial infrastructure. The deployment decision should rest on a specific agent’s current skill and the quality of the review process around it.