Real footage, called fake
On 5 November, Insurance Business reported that Zurich UK had detected a record £109.3 million of fraudulent claims in 12 months, across 4,359 claims. Scott Clayton, Zurich UK’s head of claims fraud and chair of the Insurance Fraud Bureau, said insurance fraud in the UK is becoming “more sophisticated, more organised and increasingly technology-enabled”. The report lists AI-generated evidence among the tactics. It landed in International Fraud Awareness Week, which runs to 21 November.
Zurich also said the same analytics let its claims teams settle low-risk claims more quickly and with less friction. The question is what happens to the genuine claim that doesn’t look low-risk.
In September, Insurance Post carried an investigator’s worry that genuine claim footage is being chalked off as AI. Doubt is starting to land on genuine evidence too.
The ABI’s figures, published on 6 October, put detected fraud in 2025 at £1.34bn, up 14%, with contrived incidents up 30%. In the US, the NICB said on 16 November that mentions of AI in questionable claims rose by more than 1400% between 2025 and mid-2026. The reflex is to buy a detector.
What a detector gives you
A detector gives you a score or a warning, after the fact, on a file someone else made, and it is only as good as the pixels it was handed. Natalie Randall of DAC Beachcroft points out that odd metadata and visual blemishes in documents “can also have entirely legitimate explanations, such as image enhancement, scanning or document compression”.

If a blemish can be legitimate, the handler is left with a worry and the claimant is left waiting.
The doubt tax
Once any photo can be faked, any photo can be doubted. Most of the talk is about the fakes.
Picture a driver who films a junction impact and sends the clip in. The clip is genuine. The detector returns “uncertain”, the file goes to a specialist, and the claimant waits without knowing the clip was the problem. Nobody lied. The system had no good way to tell, so it erred towards suspicion.
Handler time, a longer cycle and a complaint with no fraud finding add up to a doubt tax. It falls on exactly the claims that don’t look low-risk, and nobody is measuring it.
Ask where it came from
A detection arms race is one the honest claimant can’t win and the insurer can’t finish. Fakes get better, detectors chase, and every false positive lands on a real person.
The better question is “do we know where this came from, and when?” That is provenance: the time and place a photo or clip was captured, a record of who has handled it since, and a clear line between evidence taken at the time and evidence assembled later. It is a data problem before it is a fraud problem. A record made at the scene is much harder to argue with than a file that turned up three weeks later.
Courts are heading the same way. A US federal advisory committee has drafted, but not approved, a new evidence rule under which, once an opponent gives enough to support a finding that an item is a deepfake, the side presenting it would have to show it is more likely than not authentic. The burden of proof moves onto the person holding the evidence.
Provenance has limits, and a determined fraudster will try to get round any scheme. A real clip of a staged crash is still a real clip. It tells you where the footage came from, not whether the accident was an accident. So it sits alongside detection and investigation, not instead of them.
Where the next pound goes
Keep the detectors. They have a job on files that arrive with no history.
Put more of the next pound on the capture, before the file reaches a handler. Make a trustworthy first record the default, taken at the scene and ready to stand behind, not something a genuine claimant has to assemble later. Then a real clip is easier to defend, and a suspicious one stands out for a reason you can explain.
The test of a fraud strategy isn’t how many fakes you catch. It’s how few honest people you doubt.
Make every report count.
Tell us what your team reports and we will show you how it works in Logincident.
Contact us