The True Cost of Deepfake Fraud in Insurance: A Data-Driven Analysis

Insurance fraud is not a new problem. What’s new is the technology amplifying it.

Generative AI has made it possible to produce convincing fake images, video, documents and audio cheaply, quickly and without technical expertise. For an industry that relies on submitted evidence to assess and pay claims, that is a fundamental shift in the threat landscape.

This article looks at the financial impact of deepfake-enabled fraud on insurers. It starts from verified figures, then makes the modeling step explicit, so you can see which numbers are measured and which are assumptions.

The Baseline: How Big Is Insurance Fraud Today?

US$308.6 Billion a Year in the US

A 2022 study for the Coalition Against Insurance Fraud estimates that insurance fraud costs American consumers at least US$308.6 billion every year, about US$932 per person. The figure spans health, life, property-casualty, workers’ compensation and auto theft.

Fraud reaches policyholders through premiums. An older FBI estimate put the cost of non-health insurance fraud at US$400–700 a year per family in higher premiums. That figure dates from 2005 and is best read as an order of magnitude.

About US$45 Billion in Property-Casualty

The same 2022 study puts US property-casualty fraud at about US$45 billion a year, and the Coalition estimates that fraud occurs in about 10% of property-casualty losses (fraud statistics). This is the line most exposed to fabricated photos and documents, and it is the baseline for the model below.

Outside the US

In the UK, insurers detected £1.16 billion of fraudulent general insurance claims in 2024, and the ABI estimates that a similar amount went undetected. In Australia, Insurance Council members detected A$560 million of opportunistic motor and property fraud in 2023, with undetected fraud estimated at around A$400 million a year. Sources and detail are on our insurance fraud statistics page.

The Deepfake Accelerant

Against this backdrop, generative AI is acting as an accelerant: it makes existing fraud easier, cheaper and harder to spot.

Identity Fraud Has More Than Doubled

Sumsub’s 2024 Identity Fraud Report found that the global identity fraud rate rose from 1.10% in 2021 to 2.50% in 2024, more than doubling in three years. It named AI-driven deepfakes as one of the main trends of 2024, describing them as part of “our everyday reality”.

The data spans all industries, but banking and insurance was among the five most affected sectors.

Voice Fraud at Industrial Scale

Pindrop’s 2025 Voice Intelligence and Security Report found that deepfake fraud attempts rose more than 1,300% during the year, and synthetic-voice attacks on insurers rose 475%, the fastest of the sectors Pindrop tracks (press release).

Insurance claims often involve phone-based reporting, recorded statements and call-center interactions. Voice cloning can reproduce a voice from a short sample, such as a voicemail greeting or a social media clip, and researchers have shown that some voice authentication systems can be bypassed (University of Waterloo, 2023).

The US$25.6 Million Wake-Up Call

In February 2024, Hong Kong police disclosed one of the highest-profile deepfake fraud cases to date. A finance worker at a multinational company was tricked into transferring about US$25.6 million after a video conference in which every other participant was a deepfake of a real colleague, including the company’s CFO (CNN, February 2024). The company was later confirmed as the engineering firm Arup.

The worker had initially suspected a phishing attempt. He was convinced only after seeing and hearing people he recognized on the call.

The case was corporate fraud, not insurance fraud. But it shows the state of the art: deepfakes good enough to fool a professional in a live, interactive setting. Static claims evidence (photos, pre-recorded video, documents) is easier to fake than a live call.

Projected Growth

Deloitte’s Center for Financial Services projects that generative-AI-enabled fraud losses in the US could reach US$40 billion by 2027, up from US$12.3 billion in 2023 (Deloitte, May 2024). The projection focuses on banking and financial services and covers generative-AI-enabled fraud broadly, not deepfakes alone. Insurance shares the same exposure points: identity checks, submitted documents and remote interactions.

Modeling the Insurance-Specific Exposure

No insurer or industry body publishes a measured figure for deepfake-enabled claims fraud, and detected cases are usually handled confidentially. So the insurance-specific impact has to be modeled.

An Illustrative Model

Illustrative — assumptions, not measured deetech outcomes.

Start from the published baseline: about US$45 billion a year in US property-casualty fraud (Coalition Against Insurance Fraud, 2022).

Now assume that AI-generated or AI-altered evidence makes fraud more likely to succeed, adding a percentage to that baseline:

Scenario (assumption)	Added P&C fraud cost	Annual cost increase
Low	+5%	about US$2.3B
Middle	+10%	about US$4.5B
High	+20%	about US$9.0B

These figures cover the US P&C market only. The percentages are assumptions chosen to show sensitivity, not estimates from a study. Substitute your own portfolio’s fraud baseline and assumptions before using them.

Several factors make the middle scenario plausible rather than alarmist:

Most fraud tooling in use today scores claims data, not the authenticity of the media submitted with a claim
Digital-first claims processes offer less physical verification
The cost and skill needed to make a convincing fake keep falling

Cost Per Fraudulent Claim

The financial impact isn’t limited to the fraudulent payout. Each fraudulent claim carries associated costs:

Investigation costs: SIU time, external investigators and, increasingly, digital forensic analysis
Legal costs: claims that reach litigation before fraud is identified generate legal expenses whatever the outcome
Operational overhead: higher fraud rates mean more investigators, more tools and slower processing for legitimate claims
Reputational cost: publicized fraud cases can damage an insurer’s credibility with policyholders and regulators
Premium leakage: undetected fraud inflates loss ratios, which feed into higher premiums for all policyholders

The Consumer Impact

Insurance fraud is not a victimless crime. As undetected losses grow, the premium burden on honest policyholders grows with them.

For commercial lines the impact is direct: businesses paying higher premiums because of industry fraud rates carry a real cost, particularly in fraud-heavy lines such as commercial auto and workers’ compensation.

Where the Losses Are Concentrated

Not all lines are equally exposed. The highest-risk areas are those most reliant on digital evidence and remote assessment.

Auto Insurance

Auto claims rely heavily on damage photos, and the shift to photo-based digital assessment makes this line particularly vulnerable. Fabricated or manipulated photos can inflate repair estimates or support claims for incidents that never happened.

Property Insurance

Property claims for storm, fire, water and theft damage depend on visual evidence. AI-generated or edited images are increasingly hard to tell from genuine photos, especially at the compressed, smartphone quality typical of claims submissions.

Health and Disability Insurance

Fabricated medical records, invoices and certificates, and impersonation in remote consultations, create new attack vectors. Voice cloning adds the risk of fraudulent authorizations over the phone.

Workers’ Compensation

Fabricated injury documentation and manipulated photos or video apply to workers’ compensation, a line the Coalition already estimates at about US$34 billion a year in US fraud.

The Detection Gap

The most concerning part of deepfake fraud in insurance is the detection gap. Most fraud detection was built for pre-AI fraud: pattern matching, claims velocity analysis, database cross-referencing and manual review.

These tools remain valuable, but they were not designed to judge whether a photo, document or recording is authentic. A generated photo matches no pattern in a claims database. A fabricated document can follow the expected format exactly. People judging media by eye or ear get at least one judgment in four wrong, even in controlled studies (see our deepfake fraud statistics).

Closing this gap takes media analysis that examines the evidence itself, alongside the claims analytics insurers already run, with a human making the final call.

The Cost of Inaction

The economic choice facing insurers is straightforward:

Invest in media verification now: add analysis of submitted photos, video, audio and documents, train claims staff on the new threat, and update workflows.

Or absorb the losses later: pay for claims fraud that legacy systems were never designed to catch.

Even a few percentage points added to the industry’s existing fraud losses would be a large number. The tools used to create convincing fakes keep getting better, faster and cheaper, so the gap widens the longer it is left open.

Insurance Fraud Statistics 2026: the verified baseline figures
Insurance Fraud ROI Calculator: model the exposure for your own portfolio
Board-Level Briefing on Generative AI Fraud: the same case framed for directors

deetech™ analyzes supported images, video, documents and audio for insurers and claims teams. Performance varies by input and model release, so a proof of concept on representative data is the right way to understand your organization’s exposure. Request a demo.

Sources cited in this article: