Aviva detected 18,400 suspect insurance claims worth £233 million in 2025. That is a record, though part of the jump comes from folding in Direct Line, which Aviva acquired in July 2025. Look at Aviva's own UK general insurance business on its own, and the direction is still clearly up.
The more important story is what kind of fraud is growing. Staged collisions, the classic "crash for cash" scheme where cars are deliberately hit to generate injury and repair claims, are declining. In their place, fraudsters are fabricating digital evidence. AI tools generate convincing photos of damaged vehicles, fake accident scenes, and polished documents: repair invoices, medical reports, even police statements. A single person with a laptop can now produce a full supporting file for a high-value claim without ever leaving home.
This is not a niche criminal problem. The Insurance Fraud Bureau tracked a four-fold increase in AI-assisted fraud cases between 2022 and 2025. Admiral, another major UK insurer, reported a 71% rise in fraud during 2025, partly linked to manipulated images. The UK government estimated that eight million deepfakes would circulate in 2025, up from 500,000 in 2023. The tools to fake reality have gone from specialist to mass-market in roughly two years.
Aviva's response is to match the threat at the same scale. Its AI system works through millions of data points from current and historic claims, looking for patterns that a human reviewer working through a heavy caseload would miss. Does the damage in the photo match how the accident was described? Are the repair costs out of line with hundreds of similar jobs in the same region? Has this vehicle registration appeared in other suspicious filings? Every new claim gets cross-referenced against a deep pool of evidence.
Critically, the AI does not make the final call. It flags likely fraud cases and surfaces them for human investigators to review. This matters because over-correction is a real risk: deny too many legitimate claims and you create a different kind of crisis, one involving regulators and angry customers. The human review layer is what keeps the system honest.
The financial consequences of getting this wrong flow outward. Across the entire UK motor insurance market, fraud adds an estimated £50 to £60 to every driver's annual premium. The Association of British Insurers detected over £1.16 billion in fraudulent claims across the industry in 2024 alone. That cost does not disappear; it gets recovered through higher prices for everyone.
Beyond insurance, the underlying shift is relevant to any business that accepts documents as proof of something. Invoices, delivery confirmations, certificates, expense receipts: all of these are now cheap to fabricate convincingly. The insurance sector is just the first industry large enough and data-rich enough to build a systematic defence. For everyone else, the practical implication is that document verification processes designed for a world where faking required skill and resources are no longer adequate. Manual spot-checks are not enough when the fakes are indistinguishable at a glance.
The businesses that will handle this well are the ones that start treating verification as an ongoing process rather than a one-time check at submission. That means cross-referencing, pattern matching, and consistency testing across multiple data points, not just looking at whether a document looks official.