Coriqo
The independent AI record · banking and clinical AI

The flight recorder for your bank's AI or clinical AIindependently verifiable.

Coriqo is the tamper-evident record of the models and AI decisions a regulated institution runs on — inventory, validations, committee sign-offs, lineage, and examiner correspondence in one place. Your examiner checks the chain against a public key, so no one takes your word, or ours. We seal and verify the evidence; we never make the decision. Built for banks whose MRM function is two people and a spreadsheet — big-bank rigor, without the big-bank department.

Selling clinical AI into health systems? The same record answers a hospital AI committee under Joint Commission RUAIH and ONC HTI-1 — see the healthcare page.

Works the way your bank already works — email, calendar invites, and CSV. No new systems for your team to learn, no runtime dependency.

Zero-retention AI. Your data is never used to train models, and no AI output enters the immutable record without a named person's approval.

Buy through your cloud. Coming to AWS and Azure Marketplace — draw down committed cloud spend, lighter vendor onboarding. Listings in review; ask about early access via Private Offer.

Maps to SR 26-2FDIC / FFIECSR 11-7 (superseded)NIST AI RMFEU AI Act21 CFR Part 11

Your models are governed in a spreadsheet, a shared drive, and an email thread — until the examiner arrives and asks you to prove what happened, when, and who signed off.

InventoryA spreadsheet nobody fully trustsa live registry — import your existing inventory via CSV in minutes
Sign-offAn email thread you hope you can still findan immutable, signed validation trail
Lineage"Ask the vendor — or whoever set it up"version ancestry, vendor, and validation history on record
Exam prepWeeks of scramble before every examone button, one dated examiner package
Credibility"Trust us — the records are accurate"the examiner verifies the chain themselves with a public key. No trust in your bank — or Coriqo — required.
How it's built

Three layers. Inputs read what you already have. The core records every event. You get the deliverables.

Coriqo reads your bank's existing records into one bank-grade, append-only system of record — then turns that record into what your board and examiner actually ask for: validation trails, committee minutes, examiner packages, findings logs, and a searchable library of the documents behind them. Recorded once, verifiable forever. Deploys in your private cloud or on-premise — air-gapped supported.

Deliverable layer The outputs your board and examiner ask for · one record underneath
Shipping now

Validation & audit

Inventory, lineage, sign-off, examiner packages

Shipping now

Documents & evidence

Central library, cited cross-document search, evidence correlated to models and controls

On the roadmap

Ongoing monitoring

Threshold breaches auto-logged as governance events

On the roadmap

Board reporting

Quarterly MRM report for the board and audit committee — generated from the record

each lens reads the same graph
Core platform Built once — the structural layer every lens draws from

Model graph

Every model, version, owner, and lineage edge — relationships no single registry holds.

Append-only event log

Every approval, breach, and finding recorded as an append-only event with cryptographic integrity verification (SHA-256) — alter any record after the fact and verification fails immediately.

V1.5

Merkle attestation

Events sealed into a Merkle tree. O(log n) inclusion proofs let an examiner verify any single event without the full chain.

Identity & policy

Who can view, change, approve, retire — humans, service accounts, time-boxed examiners. Every access logged.

read-only ingest · sources keep ownership
Connector layer Read-only · pluggable · never owns the data
Registries
MLflow · Hugging Face
Storage
S3 · Azure · GCS · NAS
Pipelines
CI/CD · training jobs
Cloud
AWS · GCP · Azure
What ships today · your core deliverables

The full record — from model to sealed proof — kept by a team of two.

Most of your models are purchased and your team is small. SR 26-2 replaced SR 11-7 in April 2026 and generally excludes banks under $30 billion, so nobody hands you the standard any more: you set it and you defend it. Register every model — vendor or in-house — run it through validation, and record everything in a tamper-evident chain your examiner verifies against a public key, not your word.

01 — Register

A living inventory

Every model and version — vendor and in-house — with owner, risk tier, and lineage. Import your existing MRM spreadsheet via CSV and be current in an afternoon.

02 — Validate

A real validation workflow

Draft → review → validated → approved → in production → retired. Independent review is enforced, never the producer — the evidence of effective challenge you're paid to provide.

03 — Attest

An exam-ready trail

Every approval, monitoring breach, and examiner finding recorded in an immutable, integrity-verified audit trail. Threshold breaches auto-write a sealed governance event. Generate the examiner package on demand.

04 — Verify

Prove it. Independently.

Every event sealed into a Merkle tree. Download a proof bundle. Run one command on your own machine — no network, no Coriqo account. The math either passes or it doesn't.

Third-party & vendor models — FIS, Fiserv, Jack Henry, FICO

Most of your models are someone else's. Coriqo governs them anyway.

Vendor documentation on file. Attach the vendor's validation reports and model cards to the version they cover — hash-bound, so "which document did we rely on" has one answer.
Your gap analysis, tracked. Log your internal user-gap analysis as findings against the vendor version — with owners, due dates, and a remediation trail the examiner can walk.
Vendor review cadence. Set the next vendor due-diligence date once; Coriqo reminds the owner, flags it overdue, and surfaces it in the quarterly board report automatically.
Scope & cadence · SR 26-2 §II and §V

Two questions the guidance leaves to you. Coriqo makes you answer them on the record.

Is it a model at all? §II excludes simple arithmetic such as spreadsheet calculations, and deterministic rule-based software with no statistical, economic, or financial theory behind it. Coriqo asks three questions and stores the answers, the outcome, and the rule that produced it — a determination with a date and a name on it, rather than an assumption.
Out of scope is not ungoverned. Every out-of-scope result records what governs the asset instead: end-user-computing controls for a spreadsheet, change management for a rules engine, your installed AI rulepacks for a generative or agentic system. Nothing is filed as simply exempt.
Your revalidation interval, per model. §V sets no frequency — it says timing varies with the model's purpose, methodology, rate of change, and data limits. Coriqo holds an interval per model with the reason and the person who set it, and the overdue list shows whether that number was decided or inherited from the old 365-day default.

Both are recorded from the model's own page — the scoping questionnaire, its history, and the interval with the reason behind it. A model nobody has scoped yet says so, rather than defaulting to in scope.

Shadow-AI discovery · AWS SageMaker, Bedrock, S3 · live

An inventory is only as good as what it leaves out. Coriqo finds the models you didn't know you were running.

Read-only cloud scan. Point Coriqo at an AWS account with a read-only role. It lists the models actually running — SageMaker packages, Bedrock models, artifacts sitting in S3 — and never changes anything in your account.
Reconciled against your inventory. Everything it finds is matched to what you already govern. Whatever doesn't match is flagged ungoverned — the shadow AI running outside your MRM program.
One decision to govern it. Promote an ungoverned model into the inventory and it enters validation like any other — with a sealed record of when it was found and who brought it in.
Live demo · sample bank · illustrative data

Walk a model from submission to an examiner package.

A working prototype with illustrative data. Click through the surfaces you'll use every quarter — the inventory, the validation lifecycle, the version lineage, and the one moment it pays for itself: the examiner package.

coriqo.io / first-national-bank / inventory Live
Model inventory14 models · 31 versions · 2 reviews overdue
ModelRisk tierValidation statusOwnerNext review
credit-risk-scorer · v3
High-risk lending model. Each transition is written to the immutable audit trail.
Immutable audit trail
No transitions yet. Advance the model to begin the record.
credit-risk-scorer · lineage
Three years of version ancestry — every retrain, recalibration, and challenger promotion on record, with the datasets and commits behind each.
retrain recalibration champion promotion challenger comparison · challenger preferred v3.9 retired · 2024-Q3 v4.0 retired · 2025-Q1 v4.1 retired · 2025-Q4 v4.2 in production GBM challenger offline eval · 2026-Q1 v4.2 provenance: consumer_loan_originations_2021_2024@v5 · 2.4M rows · commit a1b2c3d4 · trained by s.obi
Generate examiner package

The moment compliance teams care about. Bundle inventory, lineage, and the full access & approval trail for a date range into a single dated document — the kind you hand an examiner.

§

Your generated examiner package will appear here.

Examination management

Not a compliance add-on. A case file built around the examiner relationship.

Other tools bolt examination tracking onto generic CRM. Coriqo treats every examination as a tamper-evident case file — the Q&A thread, findings, and remediation trail are all part of the same integrity-verified record as the approval trail.

Case identity

Every exam gets a case ID and an email alias

EXAM-2026-OCC-CR-001 becomes both a case record and an email address. Examiners reply in their normal client — replies auto-thread into the case. No portal login required on their side.

Immutable thread

Every message enters the chain on send

Messages from the examiner, responses from your team, and document submissions are each recorded as governance events. The thread cannot be edited after the fact — the chain breaks if anything is altered.

Findings & remediation

Findings become tracked governance events

An examiner raises a finding — it writes to the chain with severity, due date, and owner. Remediation closes it. The full lifecycle, from finding to resolution, is in the same tamper-evident record as the approval trail.

coriqo.io / org_risk_models / examinations / EXAM-2026-OCC-CR-001 Live
EXAM-2026-OCC-CR-001 In progress
OCC model risk examination — Q2 2026
ExaminerD. Williams · OCC
Opened14 Apr 2026
FrameworkSR 26-2
Chain events47 recorded
OCC
D. Williams · OCC Apr 14 evt_a3f2
Please provide the MDD, validation report, and last two quarters of performance monitoring. Include disparate impact testing for the retail cohort.
JP
J. Park · Model owner Apr 14 evt_b91c
Uploaded MDD v3.1, Q4 2025 & Q1 2026 validation reports. Disparate impact ratio 0.91 — within 0.80 threshold. Monitoring CSV for full history attached.
OCC
D. Williams · OCC Apr 22 evt_c77d
MDD does not include a challenger model comparison. SR 26-2 §III treats effective challenge as sound practice. Remediation plan required by May 16.
Finding raised · High · SR 26-2 §III
Reply to examiner or add an internal note… Send
All messages recorded into the governance chain on send — unalterable after the fact
A still of the examination case file. Nothing on this image is clickable — ask for a walkthrough to drive the real thing.
Committee governance & the AI co-pilot

Your committee meets like it always has. The record writes itself.

The hardest part of MRM at a smaller bank isn't the work — it's the record-keeping after the work. Coriqo removes it: the tools live inside the email and calendar workflow your committee already uses, and AI does the drafting while people keep every decision.

Meetings

Invite Coriqo to the meeting

Every committee meeting gets its own email address. Add it to the calendar invite once — the agenda pack goes out before the call, and minutes replied to the thread become a draft decision record. One habit, zero data entry.

Decisions

Minutes to chained record in minutes

AI reads the minutes and pre-fills the decision — attendees, quorum, votes, dissents, conditions. A person reviews, corrects, and records. Only the human-approved record enters the chain; conditions become tracked findings automatically.

Challenge

A system that respectfully disagrees

Coriqo continuously cross-checks your own record for contradictions — an approval sitting on an open critical finding, a sign-off after an unresolved breach — and raises the flag with the evidence attached. Accepting or dismissing it, with reasons, is itself a chained event: documented effective challenge, on demand.

Documents & evidence · shipping now

The proof behind every decision lives in a hundred PDFs. Coriqo makes it one record you can search — and hand over.

Validation reports, vendor model cards, committee decks, monitoring exports — the evidence behind every model decision arrives as documents, scattered across email and shared drives. Coriqo reads them into one place, ties each one to the model or control it supports, and keeps the whole set answerable.

Shipping now

One document library

Every document your bank uploads in one place, not scattered across email, SharePoint, and a shared drive. Open any one and read it inline.

Shipping now

Ask across every document

Ask a question in plain language and get a short answer with the exact source passage cited — not a guess. Follow the citation straight to the paragraph it came from.

Shipping now

Correlated, then confirmed by a person

Coriqo proposes which model, control, or finding each document is evidence for. A reviewer approves or rejects it in a queue. Nothing counts as evidence until a named person says so.

Shipping now

Coverage you can hand over

See which NIST AI RMF control objectives your confirmed evidence covers, and export a control-indexed pack — each item stamped with who approved it and when. Advisory, not an automated sign-off.

The trust architecture

Every other governance tool says "trust us." Coriqo lets you verify.

Certificate Transparency — the append-only log system every browser uses to police TLS certificates — proved that math-based verification scales better than institution-based trust. Coriqo brings the same, well-established audit technique to model governance. Each tier below removes one dependency on trusting Coriqo as an institution.

Tier What you must trust How it's verified
SharePoint / email The bank's IT team, the vendor, and whoever last touched the file You can't — a document can be edited silently
Coriqo V1 Coriqo's key management and servers Ed25519 signed checkpoints — but Coriqo holds the key
Coriqo V1.5 ✓ live SHA-256 only. Trust the math, not the server Merkle inclusion proof — verify any single event in milliseconds with a public key and no network connection
Coriqo V2 Any one of several independent witnesses Checkpoint roots cosigned by third-party witnesses (transparency.dev) — tamper requires compromising all witnesses simultaneously
Coriqo V3 Nothing — zero trust ZK attestations: prove "all Tier-1 models validated within 12 months" without revealing which models, who reviewed them, or their metrics
The next phase of the record · built, rolling out

SR 26-2 leaves generative and agentic AI outside its scope. Your board still has to answer for what they do.

A vendor's generative model is non-deterministic and changes without notice. An AI agent doesn't just score a loan — it can approve it, draft the letter, and file the report. When AI does the work, the examiner's question stops being "explain the model" and becomes "prove what the AI did, in order, and that no one edited the record afterward." That proof is what Coriqo holds. Footnote 3 puts generative and agentic systems outside the guidance and then hands the question straight back: a bank's own risk management and governance practices decide the controls for what the letter does not cover. Coriqo governs them against the AI rulepacks a workspace installs — NIST AI RMF and the EU AI Act — instead of stretching SR 26-2 over systems it says it does not reach. The same tamper-evident record that seals a committee decision now extends to autonomous AI decisions — a flight recorder for AI inside a regulated bank.

Inventory

Govern agents, not just models

An agent is a governed model whose mandate — what it's allowed to do — is a validated version. Every decision it makes is sealed into the same tamper-evident chain as your committee decisions: the inputs it saw, the tools it called, the evidence it cited, in order and unaltered. The agent inventory rolls it up into a control tower — how many decisions are sealed, and how many ran outside their mandate.

Assurance

Checked continuously, not once a year

A model that changes weekly can't be validated annually. Grounding checks re-run on a schedule and seal a dated result, so "it stayed grounded" is itself something the examiner can verify. A decision that uses a tool outside its mandate is flagged the moment it happens and opens a finding.

Independence

The referee a player can't own

A regulator won't accept an AI that grades its own homework. Coriqo's verifier checks any provider's output against your sealed records — model-agnostic, because the provider is only a label, never part of the verdict. Buy it through your cloud's marketplace if you like; its independence is in what it checks, not who sends the invoice.

Posture

Read-only by design

Coriqo observes and records. It never moves artifacts, never sits in your serving path, never becomes a runtime dependency.

Identity

SSO, RBAC, examiner tokens

OIDC for humans, scoped keys for pipelines, and time-boxed read-only access for examiners — every access itself logged.

Deployment

Runs in your environment

Kubernetes-native and built for air-gapped and on-prem deployment where regulated workloads require it.

Evidence

Append-only audit trail

Every state change is an append-only event with cryptographic integrity verification. Any after-the-fact alteration is detected immediately — no tamper goes unnoticed.

Trustless

Verify without trusting us

Merkle inclusion proofs. Ed25519 signed checkpoints. A public key is all an examiner needs to verify the chain on their own machine. No server, no Coriqo account.