Blog

Altered photos and AI-generated documents in auto claims

How edited and AI-made photos and paperwork show up in auto claims, how insurers test them, the legal risk, and what to do if your genuine file is questioned.

Insurers now treat every photo and document in a claim as something that can be checked, and they check with a mix of embedded data, comparisons against other claims, document forensics and old-fashioned corroboration. Submitting an edited or fabricated image or paper is a form of insurance fraud, with consequences that run from a denied claim to criminal charges. If your files are genuine, the best protection is a habit: keep the originals, send them through the insurer's own channel, and answer a question about them in writing.

This article explains the technology as a mechanism, uses dated public examples where they exist, and covers what honest policyholders should do when a real photo is wrongly doubted.

The simple edits that came first

Long before image generators, claim files held doctored evidence. The common forms are still the most common:

  • Reused damage photos. A photo from an earlier loss is submitted as new damage. A 2025 Verisk article, which carries the National Insurance Crime Bureau's copyright, says this scheme has been used for years and is likely to go unnoticed, especially when the claimant switches carriers and no record of the earlier images exists (Verisk, August 2025).
  • Downloaded images. The article describes fraudsters taking a photo of damage from the internet and submitting it, which is easier when no adjuster visits the vehicle.
  • Altered dates. A change to the date on a repair estimate, a receipt or a photo stamp, to bring an old loss into the policy period or to match a story.
  • Doctored estimates and invoices. A repair estimate edited to raise the total, a tow bill with an extra zero, a rental receipt reissued.
  • Reused photos by suppliers. The same Verisk piece describes a carrier investigation of an appraiser who submitted 170 duplicate photos over two years, affecting claims with more than $1 million in indemnity paid, and notes that repair contractors or body shops might also reuse images. Our article on body shop billing fraud: phantom repairs, parts swaps, padded labor covers the shop side.

None of this needs special software. A basic photo editor or a PDF editor is enough to change a number.

What generative tools change

Generative image tools can build a picture from a text description, or change part of a real one. The Verisk article describes two kinds of manipulation. AI can generate a deepfake image from scratch, or it can selectively alter an existing image by inpainting or filling sections to change or hide information, using apps or features now built into phone operating systems. Conventional editing software can produce what the article calls "shallowfake" images, edits realistic enough to pass a cursory look. The same article notes Adobe's "Generative AI Fill" feature in Photoshop launched in September 2023 and says scammers adopted it quickly to forge visuals for inflated claims.

Public demonstrations make the point. In February 2026, Insurance Business reported on research by the analytics company SAS that used generative tools to make three claim images: one entirely synthetic collision scene, one real car whose windscreen damage was added and whose bystanders and number plates had been altered, and one real photo (Insurance Business, February 2026). A SAS press release of May 28, 2026 describes similar altered images, including a fabricated crack and stain on furniture, and says a joint survey with the Association of Certified Fraud Examiners found only 7 percent of anti-fraud professionals thought their organizations more than moderately prepared for AI-driven fraud (SAS, May 2026). The release also mentions a short-term rental company that found a host had used manipulated images to falsely accuse a renter of causing thousands of dollars in damage.

Treat those sources with care. SAS sells fraud-detection software, and the demonstrations are staged to illustrate a point. They show what is technically possible, not how many auto claims contain such images. No public source this article relied on gives a reliable count of AI-altered auto claims.

What generative tools change in practice:

  • Skill drops. Editing no longer takes training. Prompting is enough.
  • Subtle edits become possible. A small crack added to a real photo, a plate blurred, a bystander erased, an existing dent made larger. Small changes can pass human review more easily than a fabricated scene.
  • Documents can be produced, not only edited. A repair estimate, medical bill, towing receipt or letter can be created in a believable layout. The Verisk article observes that document fraud remains a concern, though detecting it is closer to insurers' established capabilities.

How insurers test what you send

No single test is decisive, and good reviewers combine several. These are the main methods:

Method What it looks for Limits
Embedded data (metadata) The device, capture date and time, sometimes location, and the software that last saved the file Many messaging and social apps strip or rewrite this data when you share a picture, so missing data alone proves little
Reverse image checks Whether the picture appears elsewhere online, such as a listing or a news story Finds reused public images, not new fakes
Duplicate detection across claims The same photo, or a close edit, appearing in another claim, possibly with another insurer Depends on the databases a carrier can reach; see ISO ClaimSearch and how fraud analytics flag claims
Image forensics Inconsistent shadows, compression patterns, cloned regions, odd edges and text Not perfect in either direction; advanced edits can pass, and compression can create false alarms
Document forensics Altered values, reused templates, duplicated serial numbers, fields that should differ but do not, fonts and timestamps Easier on digital documents with selectable text than on scans or photos of paper
In-app capture and live photos Photos taken inside a claims or virtual-appraisal app, sometimes with prompts for specific angles Not every insurer or claim type uses it
Corroboration Police reports, shop photos, event data recorders, telematics, weather records, witness statements Slow, but hard to fake in agreement

The Verisk article says carriers extract the rich metadata embedded in digital media to identify potential modifications and to corroborate the location and date of capture with claim information, and that Verisk's digital media forensics is available through ClaimSearch. It also notes that metadata in documents can reveal what software created a file and when, and that digital signatures such as DocuSign give a way to verify documents systematically.

Two threads from our other articles matter here. Insurers use photo estimates and virtual appraisals more than they once did, which means more claims rest on images the claimant took, as explained in photo estimates and virtual appraisals: how accurate are they?. And a flag from automated screening can slow a claim even when nothing is wrong; when an automated fraud flag delays your claim describes how.

An altered photo or document in a claim can lead to separate consequences, and they can combine.

Claim denial and policy action

Personal auto policies commonly contain a fraud provision. A typical version reads that the company does not provide coverage for any insured who has made fraudulent statements or engaged in fraudulent conduct in connection with any accident or loss for which coverage is sought. Check the wording in your own policy, because forms differ. The practical result of a finding can be a denied claim, not only for the altered part but for the whole claim, and the insurer may pursue the return of money already paid. An insurer may also seek to cancel, not renew or rescind the policy. Our article on misrepresentation and rescission: can an insurer void your policy? explains when a misstatement can void coverage, and claim padding vs fraud: where the legal line falls explains how insurers and courts treat exaggeration versus fabrication.

Criminal charges

State insurance fraud statutes generally reach false claims and the documents prepared to support them. California offers a clear example: under Penal Code section 550, it is unlawful to knowingly present a false or fraudulent claim for payment of a loss, including a vehicle loss, and to knowingly prepare, make or subscribe a writing intended to be presented in support of a false claim. Those violations are punishable as felonies with prison terms of two, three or five years and fines up to $50,000 or double the amount of the fraud, and the court must order restitution. Other states have their own statutes and penalties, and our guide to penalties for auto insurance fraud gives a wider view.

Federal law can apply too. The wire fraud statute, 18 U.S.C. 1343, covers a scheme to obtain money or property by false pretenses carried out through interstate wire communication, with a maximum of 20 years. A claim submitted by online upload or email across state lines can fit that description, though federal prosecutors choose their cases and most small claims are handled at the state level.

Other consequences

The claim goes into industry databases, which can affect future coverage and prices. See Pulling your own records: CLUE, MVR and claim file. The case may be referred to the insurer's special investigations unit and the state's fraud bureau, as how insurers investigate fraud explains.

A licensed attorney is the right person for questions about specific exposure, because outcomes turn on intent, the amounts and the state.

How an honest claimant protects authenticity

You do not need to prove your photos are real in advance, but a few habits make a challenge easy to answer.

  1. Use the phone's native camera and leave the originals untouched. Do not crop, filter, retouch or apply "enhance" or beauty modes. Phone cameras already process images, which is normal; extra editing apps add changes that look like manipulation.
  2. Keep the original files. Store copies of the unmodified photos and videos on the device and in a backup, with their full resolution. Screenshots and forwarded copies lose detail and data.
  3. Check your phone's clock. Date and time stamps help only when the clock is right.
  4. Photograph in layers. Take a wide shot showing the whole vehicle and plate, mid-range shots of each damaged area and close-ups, plus the odometer and VIN plate. Include the surroundings. Our checklist, after-accident checklist and information form, and guide to a photo inventory of your car for claims cover what to capture, and dashcam footage and scene evidence: what to capture and keep covers video.
  5. Upload through the insurer's portal or app when it offers one, rather than sending pictures by a messaging app that compresses them.
  6. Do not retype or "clean up" documents. Keep the shop's original estimate or invoice as it was issued, and ask the shop to send it directly to the adjuster if you can. If a number is wrong, ask the issuer for a corrected document and keep both.
  7. Never reuse an image from another claim, a listing or the internet, even to illustrate damage you really have. Describe the damage in words and photograph your own car.
  8. Log what you sent and when. A short note in a claim diary with the file names, upload times and who received them gives you a timeline if questions arise.

What to do when a genuine file is wrongly flagged

Detection tools produce false positives. A photo can look edited because of compression, because a phone applied computational processing, or because the metadata was stripped on the way to the insurer. If the adjuster says a document or image is questioned:

  1. Stay calm and keep it factual. A question is not an accusation, and adjusters often ask simply to verify.
  2. Ask in writing what is being questioned, which file, and what the insurer needs to resolve it.
  3. Provide the originals. Send the unedited files directly from the device, or offer to let the insurer examine them in a way that protects your privacy. Offer the backup copy as well.
  4. Offer independent corroboration. The repair shop's own photos, a police report number, a tow record, a witness, dashcam video or a second appraiser all support a genuine file.
  5. Keep every exchange in writing. Confirm phone calls by email.
  6. Know the timing rules in your state. The NAIC's model claims-settlement regulation says an insurer must reply within 15 days to communications that suggest a response is expected, must accept or deny within 21 days after proof of loss, must tell the claimant within 21 days if it needs more time, and must send a letter explaining why every 45 days after that. A denial must be in writing and cite the policy provision. The model also relieves the insurer of those timing rules when it has a reasonable basis, supported by specific information, to suspect the claimant fraudulently caused or contributed to the loss, though it must still decide within a reasonable time (NAIC Model Regulation 902). States adopt their own versions, so check yours.
  7. Escalate if the claim stalls. Ask for a supervisor, then file a complaint with your state insurance department. See My claim is under SIU review: what to expect and your rights for what a formal review involves, including your rights.
  8. Consult a licensed attorney if the insurer denies the claim on authenticity grounds or asks for a sworn examination.

A script for the call or email can keep it simple:

  • "Which file or document do you want to verify, and what is the concern?"
  • "I can send the original files from my phone. How would you like me to send them?"
  • "Here are the shop's own photos and the report number. Can they be added to my file?"
  • "Please confirm in writing when you will decide, and what else you need."

Regulators are watching the tools, too

Insurers use AI to look for fraud, and regulators expect them to do it within the law. The NAIC says its Model Bulletin on the Use of Artificial Intelligence by Insurance Companies, adopted in December 2023, states that decisions or actions made or supported by AI must comply with all applicable insurance laws and regulations (NAIC). The page does not say how many states have adopted it. Our article on How regulators oversee algorithms and AI in car insurance covers what that means, and the guide How insurers use AI in pricing and claims describes how insurers use the technology in pricing and claims. For a claimant, the practical point is that an automated flag is a starting point for human review, not a final finding.

Common questions

Can an insurer tell if I edited a photo?

Often, but not always. Embedded data, forensic analysis and cross-claim comparison catch many edits, and good fakes can pass. Do not rely on an edit going unnoticed; the risk is high and the downside is a denied claim or criminal charges.

Does brightening or cropping a photo count as altering it?

Basic adjustments to make an image clearer are different from adding, removing or changing the damage shown, but any editing muddies the file. Send the original, and if you want to point at something, describe it in words or send a second marked-up copy clearly labeled as such.

What if my phone's AI features changed my photo?

Phones apply processing such as noise reduction and scene optimization to nearly every picture. That is normal and does not change what the car looks like. Features that generate or fill in image content are different; turn them off when documenting a claim.

Can I be accused if the shop altered the estimate?

You can be asked about it, since the claim is yours. Obtain the estimate directly from the shop, keep your copy, and tell the insurer promptly if you notice a discrepancy. See body shop billing fraud: phantom repairs, parts swaps, padded labor.

Do I have to hand my phone to the insurer?

Policies generally require cooperation with the investigation, but what an insurer can demand, and how you can protect private data, depends on the policy and state law. Offer the specific original files first, and talk to a licensed attorney before handing over a device.

Key takeaways

  • Photos and documents in a claim can be checked by metadata, reverse image search, duplicate detection across claims, document forensics and corroborating records, and no single method is decisive.
  • Reused photos, altered dates and doctored estimates came first; generative tools make subtle edits and invented documents cheaper, but public sources give no reliable count of how often they appear in auto claims.
  • Submitting altered evidence can mean a denied claim, policy action, state felony charges and, in some cases, federal wire fraud exposure.
  • Protect genuine claims by taking unedited photos, keeping the originals, uploading through the insurer's channel and never recycling an image.
  • If a real file is flagged, ask in writing what is questioned, send originals plus independent corroboration, track the response deadlines, and escalate to your state insurance department or an attorney if needed.

Terms used in this guide

Related states

Related guides

Related tools

Latest on this topic