Uncertainty, surfaced
Unsure results go to a person.
In development
Trust infrastructure
for the AI era
Detect forged documents, fake IDs and deepfakes before you fund a loan or pay a claim. Document fraud detection for mortgage, HELOC and insurance teams.
What you get
See what was flagged, and where.
Every finding points to the field, says how it was changed, and how strong the signal is.
Example BankStatement · Sep 1–30, 2026
J. Example · Checking ···· 4821
| Date | Description | Amount | Balance |
|---|---|---|---|
| Sep 1 | Opening balance | 3,418.60 | |
| Sep 5 | Payroll · Lakeside Logistics | +2,940.00 | 6,358.60 |
| Sep 8 | Rent · autopay | −1,850.00 | 4,508.60 |
| Sep 12 | Card purchases | −1,275.33 | 3,233.27 |
| Sep 19 | Payroll · Lakeside Logistics | +2,940.00 | 6,173.27 |
| Sep 30 | Closing balance | 14,173.27 |
- Strong
Closing balance re-rendered
Digits drawn over the original. The transactions add up to 6,173.27, not 14,173.27.
- Moderate
Deposit line overlaid
The Sep 19 text layer doesn’t match the rendered page. Font and spacing differ from the rest of the column.
- Moderate
Metadata gap
Last saved by a PDF editor three days after the statement date. Not proof on its own.
Each finding says where it is, how the file was changed and how strong the signal is. Your team decides.
API
The trust layer, as an API.
Send a document, image, video or ID. Get a verdict, its reason and every finding back as JSON.
curl https://demo.trustable.ai/v1/documents \
-H "Authorization: Bearer $TRUSTABLE_API_KEY" \
-F file=@statement.pdfimport os, requests
key = os.environ["TRUSTABLE_API_KEY"]
r = requests.post("https://demo.trustable.ai/v1/documents",
headers={"Authorization": f"Bearer {key}"},
files={"file": open("statement.pdf", "rb")})
print(r.json()["verdict"])import fs from "node:fs";
const url = "https://demo.trustable.ai/v1/documents";
const body = new FormData();
body.append("file", await fs.openAsBlob("statement.pdf"));
const res = await fetch(url, { method: "POST", body, headers: {
Authorization: `Bearer ${process.env.TRUSTABLE_API_KEY}` } });
console.log((await res.json()).verdict);Let AI agents run Trustable checks through MCP.
In developmentAdd Trustable document checks to this app. For each uploaded bank statement, pay stub, W-2 or ID, POST the file to https://demo.trustable.ai/v1/documents as the multipart field "file", with the header "Authorization: Bearer $TRUSTABLE_API_KEY". Show the returned verdict and verdict_reason to a reviewer. Never reject a file on the verdict alone: a person decides.Insurance claims
Fake damage, real payouts.
Illustrative cases, flagged before payout.
Auto · Photo
A dent that wasn’t there
Damage added with an image generator. No camera data.
Property · Photo
Storm damage, edited in
Hail damage painted onto an intact roof.
Auto · Invoice
A bigger repair bill
Total edited. One amount in a different font.
Contents · Receipt
A receipt with a new date
Purchase date and total changed.
Built for fraud, underwriting and claims teams
Mission
Make trust verifiable in every interaction. Human or AI.
The trust layer we’re building
01
Detect
Spot fakes.
02
Verify
Confirm source and identity.
03
Authorize
Set agent limits.
04
Record
Keep the evidence.
05
Compliance
Meet AI rules.
06
Copyright ©
Prove who owns it.
Detection
It starts with documents. Live today for documents, images, video and IDs.
Forged statements, pay stubs and IDs.
AI-generated or edited photos.
Face swaps and other deepfakes.
Where a file came from.
What was flagged, and where.
A trust check before agents act. In development.
Detection is just the start.
What we’re building around every check.
Cross-document checks
Income compared across pay stubs, W-2s and deposits.
In development
Agent trust check
Agents check a file before they act.
In development
Workflow routing
Pass, review or hold.
In development
Review console
All the evidence in one place.
In development
Where trust breaks
Fake documents, images, videos and IDs cost real money.
Altered statements and pay stubs inflate income.
Document
Edited income documents slip past lighter checks.
Document
Fake damage photos and invoices inflate payouts.
Image · Document
Synthetic selfies and edited IDs.
ID · Video
Edited receipts and screenshots.
Image · Document
Agents acting on fake files.
Any file
46%
of Fannie Mae’s 2024 misrepresentation findings involved income. National Mortgage News
$5.38
cost to US lenders for every $1 lost to fraud. LexisNexis, 2025
44%
of North American financial institutions rely mainly on manual fraud processes. LexisNexis, 2025
Evidence, handled
with care.
Your files, your control
We agree how files are handled before you share any.
People decide
Every flag comes with its evidence.
Security
Security built for regulated teams.
Our cloud, yours or on-prem
Host in the US, EU or APAC, deploy in your own VPC or on-prem, or keep files in your cloud while Trustable runs the checks.
SOC 2 Type II and GDPR
Independently audited controls, with personal data handled under GDPR.
FAQ
Questions, answered.
Which documents does Trustable check?
Bank statements, pay stubs, W-2s, tax returns and IDs in loan files; damage photos, repair invoices, receipts and claim documents in insurance claims; and images and video for edits, AI generation and face swaps. Each file is checked for edited text, font and layout mismatches, metadata gaps and totals that don’t add up, and every finding is marked where it sits with its strength.
How are fake pay stubs and W-2s detected?
Each pay stub and W-2 is checked for edited figures, font mismatches and rebuilt templates. Cross-checking income across a whole loan file is in development.
Why do HELOC lenders need document fraud detection?
HELOCs often close with lighter checks than a first mortgage, so an edited pay stub or statement is easier to miss.
Can Trustable detect AI-generated or edited claim photos?
Yes. It looks for edits, AI-generation traces and missing camera data in damage photos, and checks repair invoices and receipts for edits.
Does missing metadata mean a file is fake?
No. Messaging apps, screenshots and re-saves often strip it, so it’s one signal among several, never proof on its own.
Is Trustable SOC 2 compliant?
Yes. Trustable is SOC 2 Type II audited, and personal data is handled under GDPR. Files are processed in the Trustable cloud, hosted in the US, EU or APAC, or in your own VPC or on-prem; you can also keep files in your cloud while Trustable runs the checks. Data handling is agreed in writing before any file is shared, and the audit report is available on request.
Is Trustable always right?
No. No AI is 100% accurate. Results can be wrong in either direction, flagging a genuine file or missing a fake, so a person should review each flag before acting on it.
Who makes Trustable?
Trustable is a product of Reality Inc., a company based in San Francisco and Toronto that has raised $2.6M. trustable.ai is the marketing site; customers sign in to the Trustable web app at demo.trustable.ai. Trustable checks documents, images, video and IDs for edits and AI generation for mortgage, HELOC and insurance teams. Read more on the About page.