It can notice the signs, and that is all it should do on its own. An AI agent can pick up patterns that suggest someone else is controlling a customer's money, record them, and pass them to a specialist team. It must not confront anyone, freeze the account, or contact the customer through a channel the abuser may see.
What does financial abuse look like in an account?
Financial abuse is one person controlling, exploiting or taking another person's money. It takes several shapes. A partner who controls every payment and leaves the customer without access. A relative who drains an older person's savings. A carer with a power of attorney who uses it for themselves. Someone who takes out credit in the customer's name.
The signs an agent might see include: a new person suddenly added to the account or given authority; contact details changed to someone else's email or phone; large withdrawals or transfers that break years of habit; the customer's income moved into a joint account they do not appear to use; a pattern of calls where another voice answers questions for the customer; or the customer saying they are "not allowed" to spend, or that someone "checks everything".
No single sign proves anything. Many have innocent explanations. That is why the agent's role ends at noticing.
Why is acting on the signs so dangerous?
Because the wrong action can put the customer at risk. If the agent sends a message about "concerns over your account" to an email address the abuser reads, it may provoke harm. If it freezes the account, the abuser may blame the customer. If it questions the suspected abuser on a call, it warns them.
So the safety question comes before anything else: is it safe to contact the customer, and how? Specialist teams in banks are trained for this. They know how to arrange safe contact, which support organisations to suggest, and when protective account changes help rather than hurt.
What may the agent do, and what stays with a person?
| Step | The agent | Stays with a person |
|---|---|---|
| Noticing patterns | Reads account changes and activity, depth 1 | Nothing |
| Noticing conversation signs | Records what was said, verbatim | Interpreting it |
| Referring | Passes signals to the specialist team, quickly | Deciding what they mean |
| Contacting the customer about concerns | Never on its own | A specialist, by a safe route |
| Freezing or restricting the account | Never on abuse grounds | A specialist, with the customer where possible |
| Challenging a power of attorney or third party | Never | Specialist and legal teams |
| Recording the customer as at risk | Never as a conclusion | A person, under the bank's policy |
The same approach applies to spotting signs of customer vulnerability more widely. Abuse often comes with financial difficulty, because the person controlling the money leaves the customer short.
An example: a change of email and a quiet message
Take an older customer whose account has been stable for decades. In one month, the agent sees the registered email change to a new address, a relative added as a third-party user, and three transfers of 2,000 each to an account in the relative's name. None breaks a rule.
Then the customer, chatting with the agent about a missing statement, writes: "My nephew handles things now. I don't really know what's in there anymore." The agent does not ask about the nephew or the transfers. It answers the statement question, then asks, in the specialist team's approved wording, whether the customer would like to speak to someone about their account in person or by phone, and what time and number would suit. The customer gives a landline number and a morning time.
The agent sends the case to the specialist team with the account changes, the transfers and the customer's words. It does not send anything to the new email address. A specialist calls the landline the next morning.
What controls make it safe?
- Noticing only. The agent has no capability to freeze, restrict or challenge on abuse grounds.
- Safe contact first. Any follow-up uses a route the customer confirms, never contact details that have just changed.
- Verbatim records. What the customer said is recorded exactly, not paraphrased into a conclusion.
- Fast referral. Signals reach a specialist quickly, with a target response time.
- Tight access. Abuse referrals are visible only to the specialist team.
Financial abuse of individuals is a different field from market abuse surveillance, which concerns trading behaviour in financial markets. The general limits are in AI agents in banking: what they may and may not do.
Questions people ask
Can the agent tell the customer what it noticed?
Not directly. A specialist decides what to say and when, because naming the concern in the wrong moment can put the customer at risk.
What if the customer says everything is fine?
The agent respects that and does not press. The referral still goes to the specialist team, which decides whether to follow up.
Does spotting abuse involve sensitive data?
Yes. Any record suggesting a customer is at risk of abuse is highly sensitive, and the agent should share it only with the specialist team and keep it out of general notes.