Discussion

AI insurance fraud needs more than a deepfake detector, says industry veteran

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#13856

AI-generated images and documents are adding to insurers’ fraud challenge, but a detector alone cannot settle whether a claim is genuine. Decerto board member Marcin Nowak argues that the stronger approach is to combine several checks, learn from investigations and keep a person responsible for the final decision.

Watch Desk analysis

What happened

In a Forbes Technology Council article, Nowak says image detectors can help, but may lag behind improving generators and can return an inconclusive result. He recommends looking across a claim for several signals: mismatched dates, a recently registered repair shop, a possible match to an earlier claim, or a policy change shortly before a reported loss. None proves fraud on its own, he cautions.

His proposed process is to send clusters of anomalies to investigators, then feed their findings back into the system. He also recommends regular audits of how many flagged claims prove fraudulent, how many are legitimate, and how many confirmed cases were missed. The final decision on a suspicious claim, in his view, should remain with a person who can see why it was flagged.

Why it matters

A false alarm is not just a bad score on a dashboard. It can hold up an honest customer’s payment and consume an investigator’s time. Nowak’s advice is that insurers should track those costs alongside missed fraud, rather than treating a detector as a set-and-forget purchase.

He also points to the NAIC’s December 2023 model bulletin on insurers’ use of AI, which he says had been adopted in 25 jurisdictions as of April 2026. The article frames fraud detection as part of a wider question: how to govern automated systems that affect consumers.

Our read

The useful takeaway is less “find the perfect detector” than “make the whole process answerable”. Combine evidence, check what happens to cleared claims, and make sure a flag can be explained before it delays someone’s money. That is advice from an industry practitioner, not proof that one recipe will work for every insurer. Still, the focus on false positives deserves more attention than another shiny accuracy score.

Read Marcin Nowak’s Forbes Technology Council article.

What to watch

  • Whether insurers measure false positives and missed fraud alongside detection rates.
  • Whether cleared claims feed back into fraud models and rules.
  • How regulators address AI systems that can delay or deny consumer claims.

Discussion spark: Should insurers be allowed to delay a claim based on a cluster of weak signals, or should a human review be required before any delay?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.