insurance claims are deepfake fraud's next target, and insurers aren't ready

In 2024, voice cloning and synthetic-media attacks cost U.S. contact centers an estimated $12.5 billion, according to Pindrop's 2025 Voice Intelligence and Security Report — and insurers, who run some of the industry's busiest call centers, sit squarely in the blast radius. A policyholder calls in a claim. The voice on the line sounds exactly right. It isn't the policyholder at all — it's a handful of seconds of scraped audio run through a cloning model, good enough to pass the voiceprint check the call center has relied on for years.

A new attack surface hiding in plain sight

Synthetic voice attacks on insurers rose 475% in 2024, and call centers now report roughly seven suspected deepfake incidents a day — a jump of more than 1,300% year over year, per reporting on Pindrop's findings. The pattern is familiar: a fraudster pulls ten to twenty seconds of a policyholder's voice from social media, clones it, and files a claim on a vehicle or property they don't own — directing the payout to a repair shop or rental agency in on the scheme. The real policyholder often doesn't find out until a premium spike or a routine review, months later.

Ironically, the industry's own push toward convenience has widened the opening. Insurers have automated claims intake to cut costs and speed up payouts, and many still treat voice-print matching as a hardened security layer rather than one signal among many. "There's a common misconception that biometric voice authentication systems represent a hardened layer of security against synthetic speech," one fraud attorney told InsuranceNewsNet, noting that automated systems now handling the bulk of incoming claims have made the gap worse, not better.

Five ways generative AI is reshaping claims fraud

Voice cloning is only one entry point. A 2026 industry report on deepfake fraud in insurance maps out how generative AI is attacking the claims pipeline end to end:

• Voice cloning to impersonate policyholders filing claims, or claims adjusters extracting sensitive information from claimants.
• AI-generated or edited photos that fabricate — or exaggerate — property and vehicle damage well enough to pass a normal-zoom adjuster review.
• Forged documents, including police reports, medical records, and repair invoices, produced in minutes with matching formatting and letterhead.
• Deepfake video evidence, from dashcam footage to telehealth consultations used in health and workers'-comp claims.
• Synthetic identities used to open policies, build a legitimate-looking claims history, and then cash out with a large fraudulent claim.

The detection gap insurers can't ignore

Humans are bad at spotting any of this. University of Florida research found people detect audio deepfakes with only 73% accuracy — not much better than a coin flip — and video deepfakes are catching up fast as generation quality improves. Yet Signicat's research found only 22% of organizations have implemented specific measures to counter AI-driven fraud. Generic detection tools don't close the gap either: models trained on clean lab datasets often score above 95% accuracy, then fall to 50–65% once they hit real claims media — compressed, poorly lit, and shot on whatever device the claimant had in hand.

What claims and fraud leaders should do now

Waiting for a high-profile loss to force the issue is expensive. Insurers that want to get ahead of the trend should:

• Treat voice-print matching as one signal, not proof of identity — pair it with liveness and behavioral checks.
• Audit the digital claims pipeline to quantify how much evidence arrives as photos, video, audio, or documents, and where each is most exposed.
• Deploy real-time deepfake detection at claims intake and in the call center, rather than relying on manual SIU review after the fact.
• Log AI-generated fraud separately from traditional fraud so patterns can be shared across the industry.

The economics favor attackers for now: a convincing voice clone costs next to nothing to produce and takes seconds, while a manual investigation takes days. Closing that gap means putting detection at the point of contact — the call, the upload, the claim — not after the payout has gone out. Corsound AI's Deepfake Detect is built for exactly that moment, flagging cloned voices and manipulated audio and video in real time so claims teams can verify identity before a fraudulent payout is authorized, not after.

Photo: Mikhail Nilov / Pexels