PrymeIntelligence

Compliance·Answer

Can an AI agent file a suspicious activity report?

An AI agent should not file a suspicious activity report. It can assemble the evidence and draft it; the reporting officer forms the suspicion and files.

No. An AI agent should never file a suspicious activity report. It can gather the evidence, build the chronology and draft the report in the format your financial intelligence unit expects. Forming the suspicion and submitting the report stay with the money laundering reporting officer, because the duty is personal and a filed report cannot be withdrawn.

Why can an agent draft but never file?

A suspicious activity report is a firm telling the authorities that it suspects a named person of money laundering or terrorist financing. The FATF Recommendations, which national AML regimes follow, put the duty to report suspicious transactions on the institution, and most regimes give a named officer the responsibility for deciding.

Three features make filing a person's act. The decision rests on suspicion, which is a state of mind someone has to hold and defend. A filed report goes into a law enforcement database and cannot be called back. And the decision not to file is just as consequential, because failing to report can itself be an offence. The full argument is in an agent can draft a suspicious activity report but never file one.

What evidence does the agent assemble?

The value of the agent is in the file, not the form. For an escalated case it prepares:

  1. Subject details: identity, identifiers, addresses, accounts and relationships, from the customer record and due diligence file.
  2. The chronology: the transactions and events in order, with amounts, dates and counterparties.
  3. The grounds: what is unusual about the activity, compared with the customer's profile and known typologies.
  4. What was checked: screening results, adverse media, prior alerts and earlier reports, including null results.
  5. The customer's explanation, if one was obtained through normal channels, and whether evidence supports it.
  6. A draft narrative in plain language, structured as the receiving unit asks.

Each statement links to its source. The suspicious activity report drafting checklist sets out what a usable narrative needs.

What does the reporting officer keep?

Decision Why it stays with the officer
Whether there is suspicion The legal test turns on a person's state of mind
Whether to file, and when Filing and not filing both carry liability
The final wording The officer signs what the authorities receive
Submission The report is irreversible once sent
Consent or defence requests, where the regime has them Affects whether a transaction may proceed
What happens to the customer relationship Must be managed without tipping off

In Pryme's terms, submission is the kind of hard-to-undo action that sits at depth 4. For SAR filing, it is outside any compliance agent's role scope, so the verdict is refused, not held.

What does a drafted case look like?

Take a challenger bank. An investigation into a small business account finds round-sum transfers from several personal accounts, each followed within a day by cash withdrawals at different machines. The business is recorded as a café.

The agent drafts the report. The chronology covers three months and some 140 transactions. The grounds section explains why the inbound pattern does not fit card takings from a café, and why the cash withdrawals do not fit supplier payments. It lists the six personal accounts, two of which belong to customers of the same bank with their own alerts. It notes that the director gave an explanation, catering for private events, and that no invoices were supplied.

The reporting officer reads the draft, removes a sentence speculating about the source of funds that the evidence does not support, and files. The agent records the filing reference against the case and sets a review date. Neither the director nor the relationship manager is told. For the working method, see how to draft suspicious activity reports for review with an AI agent.

What controls are non-negotiable?

  • No submission capability. The agent has no connection to the reporting portal at all.
  • No speculation in drafts. Every claim in the narrative must have a source. The officer should be able to strike anything unsupported.
  • Tipping-off controls. The agent's role scope prevents it from sending anything to the subject or the subject's relationship manager about the case.
  • Restricted access. Case files and draft reports are visible only to the investigation team and the officer.
  • Decisions not to file are recorded too, with the officer's reasons.

The principle behind all of this is in our guide to AI compliance agents.

Where Pryme fits

In Pryme, drafting a report and filing one are different capabilities, and only the first exists for a compliance agent. A request to submit is refused at the role scope, and the verdict names that bound. Each draft is filed with its sources, so the reporting officer can see which facts came from the ledger and which from the customer's own account.

Questions people ask

Can the agent pre-fill the reporting portal for the officer?

Some firms let the agent prepare a file the officer uploads. That keeps the act of submission with a person while saving re-keying.

Can the agent decide a case does not need a report?

It can recommend closing an investigation with reasons. The decision not to report is the officer's, and it is recorded.

Does the agent need to know about earlier reports?

Yes, within the restricted team. Prior reports on the same subject are relevant evidence and help the officer decide.

Decide what your agents may do. Prove it. Enforce it.

Pryme Intelligence certifies each agent on your own cases, checks every action against five bounds at the moment it happens, and files every verdict.