Attestiq

AI assurance for regulated money movement

Prove your AI moves money correctly.

Attestiq checks every AI output in a payments workflow against your records and the rules that govern it, holds what is wrong for a human, and keeps evidence an examiner can recompute — before your auditor, your bank, or the OCC asks.

disputes.cardholder_letter

Reg E denial letter · draft 3

Hold
40
score / 100
3
errors caught
0
critical
  • highFACT.AMOUNT_MISMATCH

    Letter says $68.40; the transaction record says $86.40.

  • highFACT.REASON_CODE_MISMATCH

    Non-receipt dispute described under Visa 10.4; record is 13.1.

  • highDISCLOSURE.MISSING_REQUIRED

    No notice of the right to request documents relied on.

disputes-us@0.1.0evidence signed · sha256
219

labeled cases in the FinVAL-seed benchmark that gates every release

0.952

precision and recall on FinVAL-seed, deterministic mode, $0 per run

~16 ms

per validation end to end through the API, no model call

28,238

synthetic authorization attempts behind the Authorize demo

Live demo

See a verdict in seconds.

Validate an AI-drafted Reg E dispute letter, or ask why an authorization rate is dropping. Real engine, synthetic records, nothing stored.

Synthetic data. Nothing you type is stored.

What happens when you press Validate

  1. Claims are extracted: amounts, dates, reason codes, card, outcome.
  2. Each is checked against the transaction and dispute records.
  3. Reg E, Reg Z, and card-network rules are applied, each with a citation.
  4. One verdict, signed evidence for every finding, and a human decides.

How it works

The product is the audit trail. Validation is how you generate it.

  1. 01

    Validate

    Your agent calls the API, or our MCP tool, after each draft. Fact, math, policy, and disclosure validators run against your records.

  2. 02

    Cite

    Every policy finding cites the rule it rests on — Reg E §1005.11, a network reason-code table, your SOP. No citation, no finding.

  3. 03

    Decide

    High-severity outputs are held. A reviewer releases, blocks, or returns them; every override becomes a labeled example.

  4. 04

    Attest

    One click exports the examiner pack: coverage, error rates, overrides, controls, and a digest of every evidence record.

One platform

One audit trail. Modules on one contract.

A compliance officer and a payments lead see the same vendor. Every module shares the evidence chain, policy packs, console, and API — adding one is a line on your order form, not a new vendor review.

Financial services

Payments, disputes, and the AI around them — for processors, fintech programs, and sponsor banks.

Investigate

Fraud & financial crime investigations

Roadmap

What happened, and can you prove it?

Alerts, Customer 360, explainable risk scores, AI-prepared cases, and a human decision on the same audit trail.

Settle

Settlement reconciliation

Roadmap

Did the money tie out?

Settlement and payout reconciliation with exception evidence.

Verify

Caller verification

Roadmap

Is the caller who they claim?

Caller verification evidence attached to the case.

Roadmap modules are shown as direction. Each one opens only after the module before it has a paying customer.

Better together

A recommendation that moves money is an AI output too.

Authorize diagnoses why approvals leak and proposes fixes. Before a payments lead sees one, Validate checks every dollar figure against the diagnosis and every retry schedule against card-network rules — and both land in the same audit trail.

  1. 01
    AuthorizePreview

    Diagnoses decline drivers and proposes a fix

  2. 02
    ValidatePilot

    Checks figures and network retry rules

  3. 03
    Audit trailShared

    Evidence, decision, and outcome on one chain

In the demo, all five Authorize recommendations pass Validate before they are shown.

The rule the whole system follows

Code decides. Models read. Evidence proves. A human approves.

01

Code decides

Verdicts, deadlines, and dollar figures come from versioned code. A model can never set one.

02

Models read

Models extract claims they must quote, and judge issues they must cite to a rule.

03

Evidence proves

Every finding is a signed record an examiner can recompute, kept write-once.

04

A human approves

What is held waits for a person. Every decision lands on a hash-chained audit log.

Built for

The teams that answer for AI in payments.

Sponsor banks

Make validated AI a requirement across your fintech program, with one report for the whole portfolio.

Processors and PSPs

Attest the dispute and decline agents you run for merchants, and show the evidence on request.

Fintech compliance

Ship AI-drafted letters and decisions with the control your bank partner and examiner will ask for.

Eight weeks to an attested AI workload.

A fixed-fee pilot on one workload: an eval report on your historical outputs and an attestation draft for your compliance team.