Use Cases

How to Check Claim Photos for Manipulation or AI: A Checklist for Adjusters

Five checks an adjuster can run on claim photos today: metadata, error-level analysis, Content Credentials, reverse image search and the original file. What each one proves, and where verification against the RAW file takes over.

ByLumethic Team
Updated 9 min read
Share

To check a claim photo for manipulation or AI generation, run five checks in order: read the metadata, run an error-level analysis, open the Content Credentials if the file carries any, reverse-image-search the picture, and request the original file from whoever took it. The first four are indicators and take minutes. The fifth is the one that proves something, because a photograph can be verified against the RAW file it came from, and a generated image has no RAW behind it.

Every insurance claim of any size ends up as a set of photographs and an argument about them. A motor appraiser photographs a dented quarter panel, and the repair estimate follows from those frames. A property adjuster documents a flooded basement, and the settlement follows from that. Months later, a repair shop, a policyholder or a court may ask whether a photograph shows what the report says it shows, and whether it was taken when the report says it was. The photograph usually cannot answer. It is a JPEG with metadata that anyone can edit.

This guide starts with the checks, then describes what Lumethic can and cannot establish about a claim photograph, and for whom. The short version: verification depends on a RAW file, so the question that matters is who held the camera.

The Checks an Adjuster Can Run Today

  1. Read the metadata. Open the EXIF data with any viewer: the file properties dialog or ExifTool. Look at camera make and model, capture time, GPS position and the Software field. A Software tag naming an editor, a modification time earlier than the capture time, a GPS position away from the loss address, or a camera the claimant does not own are all questions to ask. Missing metadata is not one of them: claims portals, messaging apps and screenshots strip it as a matter of course.
  2. Run an error-level analysis. ELA re-saves the JPEG at a known quality and shows where the difference between the two versions is unusually high or low, which can mark regions with a different compression history from the rest of the frame. Free tools exist, FotoForensics and Forensically among them. Read the result as a hint: resized, re-compressed or heavily textured images produce patterns that look like edits and are not.
  3. Open the Content Credentials. If the file carries a C2PA manifest, an inspector shows who signed it and what it records: a phone capture, a Photoshop edit history, or a generator such as Firefly. Our AI Photo Checker reads that record in the browser. Most claim photos have no manifest, and a missing one says nothing either way.
  4. Reverse-image-search it. Google Lens, Bing Visual Search and TinEye find the same damage photograph in earlier claims, in repair-shop marketing, or in stock libraries. This catches the recycled photo, which is more common than the generated one.
  5. Request the original. Ask for the file as the camera wrote it: the RAW from a camera, or the untouched original from the phone with its metadata, plus the frames taken before and after. Then compare the submitted image against it. This is the check that turns the earlier indicators into a finding, and it is where the rest of this guide begins.

AI detection tools that score a bare JPEG exist and can serve as a sixth check. Their output is a probability, which the next section puts in context.

Who Holds the Camera

Two kinds of photograph reach a claims file, and they are not alike.

The first kind is taken by the policyholder. After a loss, the claimant photographs the damage with their own phone and uploads the pictures through the insurer's app or web portal. The portal typically strips metadata and recompresses the file. What arrives is a JPEG with no history. Lumethic can inspect such a file for content credentials and read whatever metadata survives, but it cannot verify it, because there is no RAW to compare against. Detection tools that score a bare JPEG exist. Their output is a probability, and we have written elsewhere about why a probability is a weak basis for a decision about a person's claim.

The second kind is taken by someone the insurer pays or appoints: a motor damage appraiser, a property adjuster, a surveyor, a repair partner documenting work, or the valuation photographer whose pictures underwrote a policy on a watch or a classic car. These people shoot with cameras that record RAW, or with a phone the insurer can prescribe. Their photographs decide the larger sums, and they are the photographs that get contested. This is where the Lumethic method applies.

What a RAW Comparison Establishes

A camera set to record RAW and JPEG together writes two files per shutter release. The RAW holds the sensor data before any processing; the JPEG is the camera's rendering of it. Lumethic takes both files and compares them.

The comparison establishes three things. The image originates from a physical sensor, which the noise structure and signal distribution of the RAW show and a generated image lacks. The delivered JPEG matches what that sensor recorded, so nothing was added, removed or moved after capture. And the metadata of both files agrees with itself and with the camera model, which exposes edited timestamps and stripped fields. A separate check looks for the traces a photograph of a screen or a print leaves behind. That signal is reported as evidence found, not found, or inconclusive, and it never carries a verdict on its own.

When the comparison passes, Lumethic signs the JPEG with a C2PA credential that records what was checked, when, and with which version of the method. Any C2PA-capable reader can open that record. The RAW is held in memory for the check and then discarded; it is not stored, and processing runs on infrastructure in the European Union. The data handling page describes this in the detail a procurement review needs.

None of this requires new equipment. Appraisers who already carry a camera need only enable RAW recording. The verification runs through the web upload, the Lightroom plugin or the API, whichever fits the office.

Where Verified Photographs Fit in a Claim

Motor damage appraisal is the clearest case. An appraisal report carries dozens of photographs and is routinely challenged over repair cost, the total-loss threshold and settlement on a notional repair. An appraiser who verifies the report photographs delivers a file in which every image proves its own capture. The dispute about whether a scratch was already there, or whether the photograph predates the accident, then has a document to settle it.

Property adjustment works the same way at larger sums. Staff adjusters and external adjusting firms photograph storm, fire and water damage, and the settlement rests on those frames. A credential on each photograph gives the file a record that holds when a policyholder disputes the extent of damage or a reinsurer audits the claim.

Underwriting documentation of valuables is the case with the longest memory. Jewellery, art, instruments and classic cars are insured on the basis of a valuation with photographs. When a claim arrives years later, the question is whether the item existed in the photographed condition on the policy start date. A verified photograph, signed at the time, answers it, and an independent timestamp on the credential fixes the date.

Repair documentation and surveys complete the picture. Repair partners photograph work before and after, and hail assessors, marine surveyors and construction inspectors document conditions on site. Wherever a professional holds the camera, the same method applies.

Capturing on a Phone

Not every appraiser carries a camera, and repair shops rarely do. For iPhone, Lumethic Capture records the unprocessed sensor mosaic alongside the photograph, which gives the verification the RAW it needs. The app attests the device through Apple's App Attest and signs the result on the phone. Today that attestation is reported as evidence rather than enforced as a gate, and we say so rather than imply otherwise.

An insurer can also offer Capture to policyholders on iPhone as an optional route, with faster handling of verified photographs as the incentive. Offered, not required: it covers one platform, and a mandatory extra step at the moment of a loss is not something a claims department will accept. An Android version is in development.

What the Carrier Receives

A carrier does not need to integrate anything to benefit. The credential travels with the photograph, and a claims handler can open it in any Content Credentials reader and see the verification record. The simplest first step for a carrier is a stated preference: report photographs from appointed appraisers should carry a Lumethic credential. That costs the carrier nothing and binds only the appraisers it already pays.

Carriers that want the record inside the claim system can call the API at intake and store the result with the file. The result is a set of findings with explicit boundaries, not a fraud score. An adverse finding routes the photograph to human review. Lumethic does not deny claims and is not built to. The decision stays with the handler, who now has a documented reason to look closer, or a documented reason not to.

For procurement, the points that usually matter are covered on the trust and security pages: no RAW retention, processing in the EU, deletion in line with the GDPR, and a data processing agreement on request.

What This Does Not Do

Lumethic does not score a bare JPEG. A claimant's photograph without a RAW receives inspection of its credentials and metadata, and the result says exactly that.

It does not decide claims. Findings are evidence for a person to weigh.

It does not replace an inspection. A verified photograph proves that the camera saw what the file shows. Whether the damage is consistent with the reported event remains the appraiser's judgment.

Frequently Asked Questions

Our appraisers shoot JPEG only. Does that rule this out? For those photographs, yes. Verification needs the RAW. Most cameras record RAW and JPEG together after a single menu change, and Lumethic Capture provides the equivalent on iPhone.

How large are the uploads? A camera RAW runs between 25 and 80 MB, and an appraisal has dozens of them. The upload happens once per report, and the RAW is discarded after the check. Offices with high volumes use the API or the Lightroom plugin rather than the browser.

Can the credential be read without Lumethic? Yes. It is a standard C2PA manifest, and any C2PA reader shows it. The Lumethic verification page adds the detailed findings.

Does this help with the EU AI Act? Not directly. The Act's transparency duties fall on providers and deployers of generative systems, not on insurers receiving photographs. A verified photograph is useful because it settles disputes, not because a regulation demands it.


Get Started

Appraisers and adjusters can verify a first report set on the verification platform today. Offices and carriers that want the API, or a pilot with one team, can write to hello@lumethic.com.

Related Reading

#Insurance#Appraisal#C2PA#Photo Verification#API