solutions · fintech

Wytness for fintech

Audit agent activity to a standard your regulator accepts.

YOUR RUNTIMEWYTNESS · OPERATORYour AI agentsees account identifierin clearWytness PII layermint pseudonym · sign event<PII:ACCT_*>keys never leave yousignedOperator surfaceholds pseudonym onlywytness_envelope_sigagt_merkle_entry_hashCompliance reviewerapproves where policyrequires itAPRA, MAS, FCA, OCC, FINRA, EU AI Act exportsigned export · pseudonyms only · tamper-evident

the question every regulator asks

Can you produce the evidence?

APRA CPS 230/234, the FCA's SYSC 8.1, OCC model-risk guidance, FINRA Notice 24-09, and the EU AI Act (credit scoring is high-risk under Annex III §5(b)) converge on one question: when an agent makes a decision affecting a customer or a position, can you produce the evidence? Application logs, LLM traces, and SIEM events aren't it. They're not tamper-evident, they're not tied to an agent identity, and they leak PANs into places PCI-DSS says they must never reach.

APRA CPS 234 §35APRA CPS 230 §47MAS TRM GuidelinesFCA SYSC 8.1FCA DP5/22OCC 2011-12 / SR 11-7FINRA Notice 24-09EU AI Act Annex III §5(b)

a scenario

A CPS 234 review, with and without

A Sydney neobank triages card disputes with LangChain agents — four thousand a month. At review, the APRA-regulated parent is asked to evidence fifty auto-resolutions: exact tool calls, data retrieved, decisions made, and proof no PAN left the environment.

Without Wytness that's CloudWatch, LangSmith, and Snowflake — three systems, no shared identity, PANs in all of them. With it, it's one signed Ledger export: agent, tool calls, tokenised PANs, decision, human-review flag, chain links — mapped to CPS 234 §35. An afternoon, not three weeks.

event.json
Fintech audit record

// Event identity:

id:"evt_01HX9FAB6P3N8M7Q4Y2K9V"
time:"2026-05-22T03:21:47.118Z"
agent_id:"agt_dispute_triage_v5"
source:"agt://dispute-triage-agent/v5"

// What the agent did:

tool_name:"transaction.execute"
decision:"allow"
source_format:"wytness_agt_py"Python wrapper
compliance_reviewer_id:"rev_d4f7"

// Parameters (PII tokenised before this row was sealed):

data.account_id:"<PII:ACCT_a3f9b1c4>"pseudonymised
data.card_token:"<PII:CARD_TOKEN_8e2d>"pseudonymised
data.amount_aud:142.50
data.currency:"AUD"
data.action:"auto_refund"
data.policy_version:"dispute-policy.2026.Q2"
data.human_review_required:false

// Integrity proof (3-layer):

agt_merkle_prev_hash:"sha256:c491f3a7..."chain link
agt_merkle_entry_hash:"sha256:7f3a91c4..."SHA-256
wytness_envelope_sig:"MEUCIQDQ4z8p..."Ed25519

one auto-refund from the scenario — the PAN never appears because it never left the bank

the deliverable

What you can prove

  • Which agent made which credit, KYC, or sanctions decision, down to the model version and prompt fingerprint.
  • That every decision affecting a customer was reviewable by a human before it went out, where your policy requires it.
  • That no agent retrieved or output a primary account number, full card data, or any PCI-scope field outside an approved tool path.
  • That the audit trail was tamper-evident: Ed25519 signatures and hash-chain links your auditor can independently verify.
  • That a regulator's request for a six-month transaction history of agent activity can be answered in hours, not a discovery exercise.

how the surfaces map

Three surfaces, one product.

/product/ledger

Ledger

Signed events, PCI-scope fields tokenised in your runtime, BYOS into your tenancy, Evidence Packs mapped to SOC 2 CC7.2, EU AI Act Articles 12–13, ISO 27001 A.8.15.

/product/registry

Registry

Agent, tool, and model-version inventory; drift after a model upgrade shows up as a change in tool-use distribution, not a war-room reconstruction.

/product/shield

Shield

Approval gates on credit decisions, sanctions overrides, and refunds, routed to ServiceNow GRC or the compliance Slack channel.

for your auditor

Frameworks that matter here

SOC 2 Type II

Trust services criteria for the audit pipeline itself.

ISO 27001:2022

Control set your bank's vendor risk team already speaks.

EU AI Act

Credit scoring is high-risk under Annex III §5(b).

GDPR

Article 22 covers solely-automated decisions with legal effect.

APRA CPS 230 / 234

§47 incident management, §35 incident response — control matrix in progress.

PCI-DSS v4.0

Audit-evidence support alongside a PCI-compliant processor; direct cardholder-data handling is out of scope.

MAS TRM Guidelines

Framework alignment available on request.

the boundary

What Wytness does not do

  • We are not a GRC platform. Wytness feeds evidence into your existing GRC (ServiceNow, Archer, OneTrust); it does not replace them.
  • We do not do post-trade reconciliation, settlement, or transaction monitoring. We log what your agents did; the trading and risk systems remain yours.
  • We are not a model risk management framework. We give your MRM team the evidence they need to apply OCC 2011-12 or SS1/23 to AI agents; the policy work is theirs.

Questions about your fintech audit obligations? Ask us.

We set no cookies. Sign-in and preferences use essential first-party browser storage only — no tracking, advertising, or third-party analytics. Privacy Policy