AI, fraud, identity, trust, product leadership, and whatever else deserves a closer look.
Views expressed are personal and do not represent any employer, partner, or client.
Agentic AI has not just industrialized fraud attempts. It has turned your risk model into an unlabeled dataset, one your rejections are quietly labeling for the other side, one attempt at a time.
The perimeter did not fall to an attack. It fell to compliance. Today's most effective fraud looks like perfect adherence, and the design premise of one-time onboarding trust is what needs to change.
Shufti's Deepfake Fraud Index landed this week with a number built for LinkedIn: document deepfakes projected to grow nearly 3,900 percent this year. The game has not changed. The denominator did. And the boring stat three paragraphs later is the one that should actually worry you.
For a decade the industry called it a willingness problem. It was never a willingness problem. It was a data-exposure problem wearing a willingness costume, and the infrastructure to solve it is now in the room.
Verification behaved like a bouncer for too long. Trust is not a fact you establish at a door, it is a relationship you maintain. The work of maintaining it is now a product problem, not a modeling one.
Turing thought the impostor would be the machine. Now the machine is the judge, and the human is the one pretending to be someone they are not. The threshold was designed to reduce abandonment.
A synthetic identity does not impersonate your best customer. It becomes one. Why fraud built on flawless behavior is a product problem, not a model problem.
Re-read operationally, the fable is not about honesty. It is about alert systems, finite trust budgets, and a village doing exactly the math your fraud analysts are doing right now.
For most of history, lying was costly. AI inverted the economics. A reflection on what happens to fraud, identity, and trust when the friction that protected the system disappears.
The most consequential piece of fraud regulation published anywhere this year did not happen in Washington. India's RBI is responding to real-time payment fraud architecturally, not technologically, and US fraud product leaders should be watching.
Every board conversation about deepfakes lands on the same question: can we detect them? It is the wrong question. A durable defense is a product, operational, and organizational decision long before it is a model decision.
Fraud prevention is a user experience discipline. A framework for making fraud decisions when the customer is in the room, not just the model.
Most AI fraud detection initiatives underperform not because the model is wrong, but because fraud detection is treated as a data science problem instead of a product problem.