Insurance fraud is splitting into two different problems, and most insurers are not yet equipped for the more dangerous one. That is the central argument of The State of Insurance Fraud 2026, a survey-based report from fraud-detection software provider FRISS. Rather than issue one report, the company has released it in four chapters between late August and early November 2026. The final installment, on industry collaboration, appeared November 4.
FRISS surveyed fraud practitioners across the insurance industry, but the published chapters do not say how many people responded, where they work or when the survey was fielded. Several charts also contain totals that do not add up to 100 percent. The figures below are best read as a directional snapshot of what fraud professionals are seeing, not as precise measurements. FRISS sells the kinds of detection and data-sharing tools the report recommends.
The first chapter describes the current fraud landscape. Auto insurance dominates: 71 percent of respondents named it their primary fraud challenge, compared with 16 percent for property and general lines and 11 percent for commercial lines. The most common scheme is still the ordinary one. Forty-four percent cited opportunistic exaggeration of legitimate claims, and 29 percent cited fake injury claims. Organized crime (13 percent) and rate evasion (9 percent) followed.
Asked what will drive fraud over the next two years, respondents most often pointed to economic and cost-of-living pressure (67 percent). Advances in fraud technology (60 percent) and organized fraud rings (58 percent) were close behind. A majority, 58 percent, said schemes have grown in volume or complexity. FRISS reads these results as two separate threats. One is high-volume opportunistic fraud that rises with financial stress. The other is a smaller but more sophisticated tier of organized, technology-enabled schemes. FRISS says no single strategy handles both.
The second chapter examines how far insurers have gotten with their own AI defenses. The answer is not very far. Only 9 percent of respondents said AI-based fraud detection is fully implemented. Another 36 percent are piloting or partially deployed, 36 percent are still planning and 19 percent have no plans. The most-cited barriers were budget constraints (29 percent) and data quality (27 percent), followed by lack of internal buy-in (18 percent) and regulatory or privacy concerns (16 percent). The most widely used tools are still traditional ones: automated red flags and business rules, and cross-checks against external databases, each used by 60 percent. Among adopters, 70 percent said AI has sped up fraud detection. FRISS argues the gap is organizational rather than technical, and that planning alone offers no protection.
The third chapter turns to how criminals are expected to use AI. Practitioners most often expect synthetic documents or identities (76 percent) and deepfake photos, audio or video (71 percent). Fewer cited automated large-scale application fraud (44 percent) and AI-coordinated fraud networks (33 percent). Fifty-eight percent expect deepfakes to become the dominant fraud tool in the near term. Yet only 20 percent treat document and media verification as a critical operational priority.
The chapter’s most striking finding concerns “agentic” AI, meaning software that can carry out multi-step tasks on its own instead of waiting for human direction at each step. Seventy-eight percent of respondents said such systems are somewhat or very likely to influence insurance fraud within two years. Just 2 percent said their organization is very well prepared for that.
The final chapter, also published by Insurance-Canada.ca, argues that no single insurer can keep up on its own. Sharing fraud data and alerts across insurers was the most popular form of cooperation, at 76 percent. Respondents also favored joint investigative task forces (56 percent), public-private partnerships with regulators and law enforcement (51 percent), partnerships with insurtech vendors (40 percent) and international cooperation (29 percent). Budget and resource limits (27 percent) and keeping pace with evolving tactics (24 percent) topped the list of obstacles. FRISS concludes that the industry needs a shared fraud-intelligence layer as basic infrastructure. It also warns that insurers whose internal data is in poor shape will have little to contribute to such a network.
The FRISS findings echo Verisk’s State of Insurance Fraud Study from earlier this year, which surveyed U.S. consumers and claims professionals. Both point to the same tension. AI tools are making fabricated photos and documents easy to produce. Insurers widely recognize the threat, but far fewer feel confident they can catch it, and both reports call for cross-carrier data sharing. Neither report breaks out workers’ compensation. Still, the forecasts about synthetic documents, deepfaked evidence and fake injury claims apply as much to medical reports, wage records and injury photos as to auto and property claims.