PUBLIC PRODUCT PREVIEW
COMPASS FOR BANKING · POWERED BY ARC
ARC is the versioned AI Risk & Compliance framework for banking AI use cases.
ARC gives Compass a structured control-and-test baseline for the declared use case. DATA Compass connects the bank-validated scope to measured and attributed evidence, exposes gaps and contradictions, and preserves the bank’s accountable human decision.
The result is a versioned assessment run and a concise Assessment Decision Record the bank can review and challenge.
Bank-owned decision · Catalog and ruleset are version-pinned · Run evidence links are pinned and attributed
Coverage basis
Pre-authored synthetic human judgment seeded by the demo rehearsal. Not a Compass recommendation.
ARC gives the assessment banking context.
Its versioned catalog of controls, tests, rules, and citations gives the bank a disciplined starting point—not a blank checklist.
Compass makes that framework operational.
It connects the bank-validated scope to measured and attributed evidence, while gaps, contradictions, and errors remain visible.
The bank owns the decision.
An accountable reviewer validates the scope and records the judgment, rationale, owner, and provenance.
THE OPERATING GAP
The inventory says what exists. The control library says what should apply. Evidence sits across systems, documents, and people. The decision often lands somewhere else.
Records the name, owner, and status, but not the full basis for the decision.
Defines the expected controls and tests, but needs accountable review for this use case.
Arrives from different sources, at different confidence levels, with gaps and contradictions.
Must keep the decision connected to its basis and preserve prior runs when the evidence changes.
When challenge arrives, the bank should not have to reconstruct the decision from email, slides, and memory.
THE RECORD
This illustrative product-preview summary uses a fully synthetic bank and use case. It shows answered tests, unresolved work, and the human decision without collapsing them into one reassuring score.
Recorded basis
Coverage state
22 of 40 in-scope tests answered
14 measured · 8 attested
18 unresolved in this completed run
2 need a person · 16 need evidence
| Outcome | Count |
|---|---|
| Satisfied | 13 |
| Marginal | 6 |
| Not satisfied | 3 |
| Awaiting a person | 2 |
| Not assessable | 16 |
| Collector errors | 0 |
Illustrative synthetic sample. “Conditionally authorized” is a pre-authored synthetic human judgment seeded by the demo rehearsal. It is not a Compass recommendation, customer result, or legal conclusion.
HOW IT WORKS
Record the intended use, owner, scope, and material dependencies.
ARC supplies the versioned control-and-test catalog. For the current Credit Scoring preview, Compass uses deterministic rules to suggest a baseline; an accountable reviewer adjusts and validates the final scope.
Bring together measured data and attributed evidence while preserving gaps, contradictions, unknowns, and errors.
A named bank user records the judgment and rationale and names the accountable owner.
When evidence or validated scope changes, a user starts a new run and Compass retains the prior one.
WHY IT IS DEFENSIBLE
In the product, Compass keeps the use case, validated ARC scope, evidence, findings, reviewer actions, positions, and human decision connected across immutable runs. The one-page Assessment Decision Record is deliberately concise: judgment, rationale and owner; coverage and outcome counts; an evidence-register summary; supersession count; and pinned provenance. A fluent explanation never replaces the underlying record.
Preserve the ARC catalog, ruleset, digest, and citations used for the run.
Keep measured, attested, missing, and failed evidence visibly distinct.
Record the decision owner, rationale, concurrence or dissent, and supersession.
A user adds a new assessment run when the basis changes instead of rewriting the prior decision.
Compass supports and preserves the decision. The bank owns it.
REGULATORY CONTEXT
In April 2026, the Federal Reserve, OCC, and FDIC issued revised, risk-based model-risk guidance that supersedes SR 11-7 and SR 21-8. It is expected to be most relevant to banking organizations with more than $30 billion in total assets, but may also be relevant to smaller organizations with significant model-risk exposure.
The guidance excludes generative and agentic AI models from its scope while directing banks to use their own risk-management and governance practices for systems outside it. Its principles continue to apply to traditional statistical and quantitative models and non-generative, non-agentic AI models. Separate interagency guidance says using a third party does not diminish a bank's responsibility to operate in a safe and sound manner and comply with applicable laws and regulations.
SR 26-2 excludes deterministic rule-based processes from its model definition when no statistical, economic, or financial theory underpins their design or use. The bank determines the classification and controls for its deployment.
Compass does not interpret the rules or accept risk for the bank. It preserves the evidence and accountable decision behind each use case.
DEPLOYMENT
DATA Compass deploys inside the customer environment and assesses connected sources in read-only mode. AI-assisted features use an in-environment model or a customer-approved external model endpoint.
We will walk one declared use case through the ARC control-and-test baseline, the Compass evidence record, and the bank’s human decision using an illustrative banking dataset. No source data is required for the first conversation.