JSJOHAN STRYDOMSubscribe free
The Actuary Lens

Validate the facts. Protect the honest customer.

Independent evidence helps resolve inconsistencies fairly.

The executive takeaway

New deception methods increase the need for proportionate validation.

From an inconsistency to a fair decision

Validation process illustration. No conversion or detection rate is implied.

Proposal, policy change or claim

Information is submitted and a check raises a question.

Corroborate the facts and explanations

Check evidence quality, recording errors and innocent explanations.

After corroboration

Clear

The concern is explained. Continue appropriately.

Correct

Resolve an error using verified information.

Investigate

A material concern remains. Use proportionate human review.

Measure confirmed value alongside false positives and the effect on honest customers.

Keep in mindA mismatch is not proof of fraud; detected fraud is not total prevalence.

What supports this view?

Established: This evidence set distinguishes reported fraud, warning signals and validation tools.

Evidence to change the assessment: Reassess with measured detection value, false positives and customer outcomes.

Changing answers is a validation problem before it is an AI claim

Evidence item dated 2026-09-22 · AI attribution not established · UK · Product documentation as at September 2026

Percayso documents comparison of quotes and retrospective policy analysis to identify manipulation.

Limit: Compare declarations across the journey and refer material inconsistencies. A changed answer alone does not establish intent or AI involvement.

  • Percayso Inform: Inform

    Evidence period: Product documentation accessed 22 September 2026Checked in this release: 22 Sept 2026

Images and documents need their own integrity checks

Evidence item dated 2026-09-22 · Supplier-described capability · International; UK validation required · Product documentation as at September 2026

FRISS describes checks for synthetic images, altered PDFs and reused evidence.

Limit: Combine media analysis with original files, incident evidence and human review. A synthetic-content score is not a fraud verdict.

  • FRISS: Media Check

    Evidence period: Product documentation accessed 22 September 2026Checked in this release: 22 Sept 2026

Evidence dates & releaseEvidence published 4 Oct 2026
Evidence period
Periods vary by sourceThe time measured. Publication and event dates appear with the individual sources.
Assessment changed
Change date unavailable in this releaseLast recorded material change to the insurance assessment.
Explainer changed
Change date unavailable in this releaseLast recorded change to its evidence or explanation. A routine source check does not reset it.
Source checks
21 Sept 2026 – 28 Sept 202640 of 40 sources have a check date in this release. These dates describe inspection, not new evidence.

Approved evidence published 4 Oct 2026This topic, its downloads and linked Data Lab views use the same selected release.

Inspect the sources and their dates
Read the explanation, evidence and limitations

Validate the facts. Protect the honest customer.

Proposal and claim information can be checked against independent evidence. The aim is to find meaningful inconsistencies while keeping fair treatment and human judgement.

Established: This evidence set distinguishes reported fraud, warning signals and validation tools.

Developing: AI can support checking and can also make false evidence more convincing.

Still uncertain: A mismatch is not proof of fraud; detected fraud is not total prevalence.

What changed in the evidence?

Changing answers is a validation problem before it is an AI claimEvidence item dated 2026-09-22 · AI attribution not established · UK · Product documentation as at September 2026

Percayso documents comparison of quotes and retrospective policy analysis to identify manipulation.

Limit: Compare declarations across the journey and refer material inconsistencies. A changed answer alone does not establish intent or AI involvement.

  • Percayso Inform: Inform

    Evidence period: Product documentation accessed 22 September 2026Checked in this release: 22 Sept 2026

Images and documents need their own integrity checksEvidence item dated 2026-09-22 · Supplier-described capability · International; UK validation required · Product documentation as at September 2026

FRISS describes checks for synthetic images, altered PDFs and reused evidence.

Limit: Combine media analysis with original files, incident evidence and human review. A synthetic-content score is not a fraud verdict.

  • FRISS: Media Check

    Evidence period: Product documentation accessed 22 September 2026Checked in this release: 22 Sept 2026

Application and claim evidence in the same review workflowEvidence item dated 2026-09-22 · Supplier-described capability · UK · Product documentation as at September 2026

Guardian Integrity Shield describes application, claims and document checks, including AI-generated evidence.

Limit: Test the extra cases found after existing controls, and the impact on genuine customers. Supplier claims are not observed performance.

Insurance implication

New deception methods increase the need for proportionate validation.

Current response: Test

Watch next: Reassess with measured detection value, false positives and customer outcomes.

A mismatch is not proof of fraud; detected fraud is not total prevalence.

Released sources

Download the underlying released evidence