It can take the report, and it should, because speed matters most in the first hour. An AI agent can record what happened, protect the account, and alert the receiving bank through the standard process. It does not decide reimbursement, judge whether the customer was careless, or close the case. Those stay with a named person.
Why does the first hour matter so much?
Money sent to a scammer moves quickly. Once it lands in a receiving account, it is often split and forwarded within minutes. The sooner the sending bank alerts the receiving bank, the better the chance that some of it is still there to freeze.
That makes intake a task where an agent helps. It is available at three in the morning and asks every question in the same order, so the person reviewing the case starts with a complete record.
What decides how much the agent can do?
The deciding factor is whether the scam is over. A customer who reports a payment they made yesterday, and who now knows it was a scam, needs a calm intake and fast protection. A customer who is still on the phone to the "bank's fraud team", or who is about to send a second payment, is still under the scammer's influence. That second case needs a person now, not a form.
The agent also watches for vulnerability: distress, confusion, a third party speaking for the customer, or a customer who has lost money they cannot afford to lose. Any of these moves the case to a person straight away.
What may the agent do, and what stays with a person?
| Step | The agent | Stays with a person |
|---|---|---|
| Taking the account of events | Asks the bank's set questions and records answers verbatim | Re-interviewing where the account is unclear |
| Protecting the account | Freezes cards and blocks new payees at depth 3 | Lifting those blocks |
| Alerting the receiving bank | Submits the standard recall or fraud notice at depth 3 | Following up on partial recoveries |
| Checking for further risk | Lists recent payments and payees for the customer to review | Deciding which are part of the scam |
| Explaining what happens next | Gives the process and timings from the policy library | Promising any outcome |
| Reimbursement | Never decides | A named person, under the bank's policy |
| Ongoing scam or vulnerable customer | Hands over at once | Taking the call |
Reimbursement is a judgement about liability, consumer standards and the customer's circumstances. We set out why in scam reimbursement needs a person and in whether an AI agent can decide whether to reimburse a fraud victim.
An example: a fake delivery fee that became a safe-account scam
Take a customer who messages the app on a Sunday evening. Two days earlier they paid a small delivery fee through a link in a text. That morning, someone claiming to be from the bank called, said their account was compromised, and talked them into moving 4,000 to a "safe account".
The agent first asks whether the caller is still in touch. The caller has promised to ring back at eight. The agent tells them plainly that the bank never asks customers to move money to protect it, freezes the card used for the delivery fee, blocks new payees, and submits a fraud notice to the receiving bank. It then routes the case to the fraud team as urgent, flagged because the scammer plans to make contact again.
A person calls the customer within the hour, before the scammer does. The record holds the customer's own words, the time of each protective step and the notice reference.
What controls make it safe?
- Questions from the policy library. The intake questions are set by the fraud team, not improvised, so every report can be compared with the next.
- Protective actions only. The agent's tools freeze, block and notify. It has no tool to refund, reverse or credit the account.
- Live-scam detection. Signs that the scammer is still in contact trigger an immediate handover, tested during certification.
- No blame in the agent's words. The agent does not comment on whether the customer should have spotted the scam. That judgement belongs to the person deciding the claim. The full intake process is in handling scam reports from customers with an AI agent, and prevention before the payment is covered in whether an AI agent can stop an authorised push payment scam. The general limits sit in AI agents in banking: what they may and may not do.
Where Pryme fits
In Pryme, the scam-intake agent is certified on the bank's own past reports. The escalation gate in Attest checks that it hands over live scams and vulnerable customers every time; a miss fails the run, which keeps only the highest depth it cleared, or drops the agent to depth 1 the same day if it cleared none. Freezing and notifying run at depth 3. Reimbursement does not exist in the agent's role scope.
Questions people ask
Can the agent tell the customer whether they will get their money back?
No. It can explain the bank's process and the usual timeline, but any statement about the outcome is a promise the bank may not be able to keep. The agent should say who will decide and when the customer will hear.
Should the agent report the scam to the police?
The agent can tell the customer how to report it and record that it did so. Whether the bank itself reports, and to whom, follows the bank's own procedures and stays with the fraud team.
What if the customer is not sure it was a scam?
The agent records the report as a suspected scam and still takes the protective steps, because they are reversible. A person reviews the case and decides how to classify it.