Agentic fraud

Fraud carried out by AI agents that plan and run the whole scam themselves, from first contact through cash-out, with little or no human directing it in real time.

Instead of a person writing each message and making each call, software does the contacting, the persuading, and the follow-up.

Fraud in which autonomous AI agents perform some or all steps in an attack chain, contacting a target, personalizing the approach, sustaining a conversation, adjusting after a rejection, without a human operator running each step manually.

An agentic fraud operation can automate several stages:

• Targeting, by picking victims and gathering details about them.
• Contact and personalization, by writing messages or speaking in a way tailored to each person.
• Conversation, by keeping a dialogue going over days or weeks.
• Adaptation, by changing approach when a victim hesitates or refuses.
• Cash-out, by guiding the victim to send money or share credentials.

The approach differs from earlier scams in a few ways:

1. Scale, since one operator can run many conversations at once.
2. Persistence, since an agent does not tire or lose track.
3. Personalization, since each message can use details about the target.

Agentic fraud sits next to terms that are easy to mix up:

• Agentic Identity is the verification layer for legitimate agents, while agentic fraud describes agents used for crime.
• Know Your Agent (KYA) is the practice of verifying agents before they act.
• Deepfake is synthetic media that agents can use to make a scam more convincing.

For more, see How AI agents are fueling a record surge in fraud losses and The Agentic Fraud Report.