Account takeover is a fraud in which someone other than the customer gains control of the customer's existing account, usually by stealing credentials, intercepting one-time codes or persuading staff to change contact details, and then uses it as if they were the customer. The account is real and the customer is real; the person acting is not.
Why does account takeover matter once AI agents act?
For two opposite reasons. First, agents become a new route in. A service agent that can change an email address, reset online access or add a payee holds exactly the capabilities a fraudster wants to talk their way through.
Second, agents become part of the defence. Takeovers leave a trail across logins, device changes and payment behaviour that is slow for people to assemble by hand. An agent can gather it in minutes. What it should not do is decide the outcome alone, because both wrong answers hurt: locking out a real customer, or releasing funds to a fraudster. The wider rules for what an agent may do in a bank are in AI agents in banking.
How does an account takeover usually unfold?
- Credentials or codes are obtained, through phishing, data from another breach, or a hijacked phone number.
- The fraudster changes contact details, so alerts go to them rather than the customer.
- They add a new payee or device, or raise limits.
- Money leaves, often quickly and in several payments.
The middle steps are where defence pays. A change of phone number followed within hours by a new payee is a strong signal; either step alone rarely is.
What may an agent do, and what stays with a person?
An agent working an alert can read the event history, build a timeline, compare device and location data with the customer's past pattern, and draft a case summary. Where the business has approved it in advance, it can place a temporary hold on outgoing payments, because a hold is reversible. The method is set out in how to investigate account takeover cases with an AI agent.
Confirming fraud, reimbursing the customer, restoring access and deciding whether to end the relationship stay with a person. Closure in particular is hard to walk back, as irreversibility in account closures and can an AI agent close a bank account? explain.
What does that look like in an example?
A digital bank's fraud agent sees a password reset, a new phone number and a new payee on one account inside two hours, then an attempted transfer of 2,000. The payee details arrived in the session, not from any verified source. The agent holds the transfer, blocks further new payees pending review, and writes a timeline for the fraud analyst. The analyst calls the customer on the number held before the change. The customer did none of it.
How is account takeover different from authorised push payment fraud?
In account takeover the customer did not make the payment; someone else did, using the customer's access. In authorised push payment fraud the customer makes the payment themselves, having been deceived about who they are paying or why. The distinction changes the evidence you need and the agent's role: in a takeover it looks for signs the customer is not the one acting; in a scam it looks for signs the customer is being manipulated. Neither is the same as a payment keyed to the wrong account by mistake, covered in how to handle payments sent to the wrong account.
Questions people ask
Can an AI agent itself be the target of a takeover attempt?
Yes. A fraudster may try to persuade a service agent to change contact details or reset access. Those changes should require strong authentication completed by the customer, and the agent should never accept a phone number or email supplied in the conversation as proof of identity.
Should an agent restore access after a confirmed takeover?
Not on its own. A mistake hands the account straight back to the fraudster, so a person should decide and record the basis.