Extending Unit21’s AI Investigation Agents from triaging alerts, investigating cases, and drafting narratives to building custom tasks, in minutes
SAN FRANCISCO--(BUSINESS WIRE)--Unit21, the leading AI Risk Infrastructure platform for fraud prevention and AML monitoring, today announced Agentic Task Builder, a new capability that lets any fraud or compliance team use AI to build the exact investigation task their program needs—in plain English, in minutes.
Every fraud and compliance program is different—the typologies it faces, the thresholds it sets, the checks analysts run before closing a case. Yet most AI tools ship a fixed set of behaviors and force teams into a template. Changing what the AI does means filing a ticket with engineering or waiting on a vendor roadmap. The Agentic Task Builder closes that gap.
Built on top of Unit21's existing AI Risk Infrastructure, which already triages alerts, investigates cases, and drafts regulatory narratives, the Agentic Task Builder removes the last major bottleneck in risk and compliance automation, putting the power to extend AI logic directly in the hands of the people running the program. Teams describe a task the way they would explain it to a new analyst, and Unit21's AI builds it, suggests recommendations to help you iterate on it, tests it against real historical alerts, and deploys it as a component of the AI Agent workflow.
One builder. Full investigation coverage.
The Agentic Task Builder produces two components: one that searches external data and resources, and one that handles internal data, both designed to cover every stage of an investigation.
-
Online search. Teams build web-research tasks powered by generative AI. Users type a plain-English prompt or select from a template, and the agent researches the public internet, returning a structured answer with citations.
Example: "Search for adverse media or sanctions exposure tied to this entity and any connected business; give a clear verdict with sources." -
Data analysis. Analysts ask questions about the alert's own data, such as transaction patterns, entity behavior, or counterparty activity, and receive a structured, AI-generated analysis directly in the alert's summary panel.
Examples: "Review the transactions in this alert for deposit and withdrawal patterns just under reporting thresholds,” or “Find me wire fraud.” It can be as vague as that.
Both task types become building blocks of the AI Agent workflow. The agent's final narrative combines all task outputs into a coherent investigation picture. Every run is explainable, auditable, and human-approved, meeting the evidentiary standards regulators require.
Agentic from the first prompt
The Agentic Task Builder doesn't just take an instruction and run it. It reasons over your environment the way an experienced analyst would. When you describe a task, the builder looks across the data already in your Unit21 environment to understand what a task like this should actually do, and asks questions back to clarify your intent before it commits.
As it works, it runs an in-built quality judge over its own output, loops on itself to find where it can improve, and recommends changes to its own prompting logic. It also suggests other data worth pulling in, surfaced from what it finds across your environment, that could make the task sharper. The result is a task you shaped together, not one you had to specify perfectly on the first try.
The analyst with AI by their side
For the investigator working the queue, the Agentic Task Builder changes what a single person can do in a day.
- A fraud ops manager can build a task that scans an alert's transactions for account-takeover and mule signals: rapid fan-in and fan-out, or new-device activity ahead of a large payout.
- An AML investigator can build a task that checks an entity and any connected business for adverse media or sanctions exposure, and another that reads transactions for structuring patterns just under reporting thresholds.
- A compliance team can deploy custom name-matching and pass-through analysis the moment their program needs it, without filing a request, without an engineering sprint.
Tasks are tuned to each program's thresholds, narrative format, escalation rules, and review controls. You define the job once, in plain English, and the agent runs it consistently across teams and geographies. You don't ask analysts to "prompt better." You define the job once, and the agent handles the rest. The analyst who shapes their own AI investigation task, tuned to the exact typologies their program faces, isn't just more efficient. They're operating at a different level entirely.
Built for self-service. Designed for trust.
Before any task is saved, teams can backtest it against historical alerts to compare outputs and validate behavior before deploying. The platform flags unrepresentative samples and surfaces warnings rather than returning false confidence, so teams deploy with evidence, not assumptions. Once a task is live, teams can sample results by where analysts and the agent agree and disagree, broken out per queue and per agent, and set up progressive autonomy at the queue level, expanding what the agent runs on its own as it earns trust.
"The best compliance teams of the future won't be the biggest—they'll be the ones where every analyst has AI built to their program's exact risks, running alongside them on every investigation. The Agentic Task Builder is how we make that real: describe the task you need in plain English, and the agent runs it consistently, explainably, and at a scale no team could match on their own,” said Tyler Allen, CEO, Unit21. “The analyst with the right AI by their side doesn't just work faster—they see more, catch more, and make a meaningful dent in the growing financial crime problem.”
This launch follows strong customer demand for Unit21's AI Risk Infrastructure, with AI Agents having reviewed over 1.5 million alerts and counting. Teams are already building custom tasks in minutes, tasks that previously required engineering hours or lengthy vendor scoping cycles.
The Agentic Task Builder is now available for customers within Unit21’s AI Agent workflow. If you’d like to see the Agentic Tasks Builder live in a demo, join us on July 22nd.
About Unit21
Unit21 is the leader in AI Risk Infrastructure, trusted by over 200 customers across 90 countries, including Sallie Mae, Chime, Intuit, and GreenDot. The platform unifies fraud and AML detection, investigation, and regulatory filing with AI agents that execute investigations end-to-end, gathering evidence, drafting narratives, and filing reports so teams can scale safely without expanding headcount. Unit21’s AI agents have processed over 1.5 million alerts to date, and its Fraud Consortium covers over 100M of the U.S. adult population across a network of fintechs, banks, and crypto platforms. Every AI decision is explainable, auditable, and defensible to regulators. Unit21 is backed by Tiger Global Management, Gradient Ventures (Google’s AI-focused venture fund), ICONIQ Capital, and others. Learn more at unit21.ai.
Contacts
Media Contacts
Unit21, Cassie Pallesen, cassie.pallesen@unit21.ai
