Impersonation fraud is a scam in which a criminal pretends to be someone the victim trusts, such as a bank, a senior manager, a supplier, a public official or a customer, to get money, information or access. The victim acts willingly, believing the request is genuine. Its defence is checking identity through a channel the fraudster does not control.
Why does impersonation fraud matter for AI agent work?
Two things have changed. First, the fraudster's tools are better: messages are fluent and personalised, and voices and documents can be faked convincingly. Second, the first reader of many requests is now an agent. An agent that reads an email from "the finance director" asking for an urgent payment, and acts on it, is an impersonation fraud victim at machine speed.
That is why governed agents treat sources differently. A fact stated in an email is external: it can open a case, but it cannot on its own authorise a payment. The people who review agent work are the next line, and employers in fraud, payments and finance operations want people who recognise impersonation on sight. Interview practice for that work is in interview answers about AI agents for fraud analysts.
How does impersonation fraud work in practice?
Most cases follow the same pattern:
- Pretext: the fraudster claims an identity and a reason, often using real names gathered from public sources.
- Urgency or authority: a deadline, a senior name, a threat of consequences.
- A changed detail: new bank details, a new phone number, a request to use a different channel.
- Isolation: "keep this confidential", "do not call the usual number".
The changed detail is where it can be caught. Any request that changes where money goes or who can be contacted should be confirmed using contact details already on file, never those in the request.
What is an example?
An agent at a manufacturer receives an email, apparently from a long-standing supplier, saying its bank has changed and attaching an updated invoice for 18,000. The agent drafts the supplier update and the payment, and both are held for a person because the new details came from an external source. A reviewer we will call Jonas notices the sender's domain differs by one letter from the supplier's real one. He calls the supplier on the number in the master file. The supplier has not changed banks. Jonas rejects both drafts, records why, and reports the attempt.
How is impersonation fraud different from account takeover?
In impersonation fraud, the criminal pretends to be someone and persuades the victim to act. In account takeover, the criminal gains control of a real account, such as a customer's login or a supplier's email box, and acts from inside it. Account takeover is harder to spot because the request really does come from the right address. The same defence applies to both: confirm changes through a separate, known channel.
Which terms are related?
- Pressure tactic: the urgency and authority that impersonation relies on.
- Social engineering: manipulating people into breaking normal procedures.
- Prompt injection: hidden instructions aimed at an agent, often part of the same attack.
If you are a fraud analyst planning your next move, see what AI skills a fraud analyst should learn to stay employable and, for one market, AI operations careers for fraud analysts in Germany. For the wider set of skills, read AI skills for job seekers.
Questions people ask
How do I show impersonation fraud skills on a CV?
Describe a catch: the signal you noticed, the check you made and the outcome, in round figures. Guidance for presenting this is in how a fraud analyst should show AI agent skills on a CV.
Can an AI agent spot impersonation better than a person?
An agent can check sender details, compare against records and flag mismatches consistently. It should not be the one to approve a change of payment details on that basis alone. The final call on anything that moves money belongs with a person who has confirmed it independently.