We don't make your filing. We make the OCC's expectations enforceable, action by action.
The U.S. Office of the Comptroller of the Currency sets supervisory expectations for a bank deploying consequential AI. The KYE OCC Filing Readiness Pack™ turns the action-boundary parts of those expectations into runtime controls: every consequential AI action proceeds only inside the use-authority you approved for it, only within the authority you recorded for that AI, under a named accountable officer — and emits a signed record an OCC examiner can replay.
Supervisory guidance tells you what good looks like. It does not stop a bad action at the moment it happens.
- Approved use is a record on paper. Your model risk management program approves a model for a use — but a policy document does not gate the action when it runs.
- The AI inventory says what exists, not what is allowed. Knowing an AI is in use is not the same as enforcing what it is authorised to do.
- Third-party and vendor AI widen the surface. An external model acting inside your boundary needs the same authority discipline as your own, per the OCC's third-party risk expectations.
- A new activity needs examiner-ready evidence. When an OCC examiner asks why an action proceeded, reconstructed notes are contestable.
- Accountability must attach to a person. Under the OCC's heightened standards, a consequential AI action that no named officer stands behind is a gap, not a control.
A signed Evidence Pack™ on every consequential action — verifiable from public keys alone.
- Use-authority enforced, not assumed. A consequential AI action cannot reach finality outside the approved use-authority you recorded for that model.
- An AI action-authority inventory. Every AI is recorded with what it is authorised to do; an action outside that authority is refused at the boundary — and the inventory is examiner-ready.
- Escalation before finality. An out-of-scope or out-of-authority action returns REQUIRE_APPROVAL and routes to the named accountable officer before it can become final.
- External AI under the same discipline. Vendor and third-party AI act only inside the authority recorded for them, per the OCC's third-party risk expectations.
- Replay-provable for an OCC examiner. Every consequential action emits a signed provenance pin binding the model, the inputs, the use-authority, and the named authority — replayable offline.
Governance at the moment of the action — across every consequential AI decision.
Every consequential AI action flows through the same gate. The pipeline binds to canonical KYE Protocol™ envelopes — no bespoke shapes per bank.
- 1 — Action proposed. An AI agent proposes a consequential action — an approval, a payment, a trade, a credit decision, a customer communication.
- 2 — Use-authority gate. The gate checks the action against the approved use-authority for that model. No resolved use-authority, no action.
- 3 — Authority check. The action is checked against the AI's recorded action-authority — including for external and vendor AI. Outside authority, it is refused.
- 4 — Escalation where needed. An out-of-scope action returns REQUIRE_APPROVAL and routes dual-channel to the named accountable officer before finality.
- 5 — Named accountability. The action is bound to a named accountable officer; accountability never transfers to the AI.
- 6 — Evidence Pack™ sealed. The action emits a signed provenance pin binding the model, the inputs, the use-authority, and the named-authority outcome, WORM-retained.
- 7 — Replay for an OCC examiner. The Evidence Pack™ is verifiable offline from public keys alone, and a change to an AI's authority is itself an evidenced, replayable event.
Bound to the OCC action-boundary subset.
The Pack binds the canonical KYE™ artefact set to the action-boundary parts of the OCC's supervisory expectations. Every claim resolves to a control row on the bound spine. We operationalise the OCC's expectations — we do not run your model risk management program, validate models, make your filing, or judge whether the AI's output is correct.
| OCC area | Control area | Pack coverage |
|---|---|---|
| Model risk management | Approved-use authority + model-change as a named-authority decision | full |
| Third-party / vendor AI risk | External authority register + escalation before finality | full / partial |
| New-activity / examiner readiness | Replay-provable Evidence Packs + action-authority inventory | full |
| Heightened-standards governance | Named accountability at the action boundary | full |
| Model development & validation | Quantitative model build, validation, supervisory determinations | out of scope |
Full coverage means every action-boundary requirement in the control area is bound to a canonical KYE™ rule, dictionary term, or Evidence Pack™ contract. Partial coverage means the bound surface satisfies the control area when paired with the bank's own attestation. Out of scope means the OCC area is the bank's own work or the regulator's own work — developing and validating models, running the MRM program, making the filing, and the OCC's supervisory determinations all stay yours (honest scope).
Run a pilot on synthetic data — then deploy with confidence.
The OCC Filing Readiness Pack™ ships in three tiers: a fixed-scope examiner-readiness diagnostic, a full pilot on synthetic data, and an annual enterprise entitlement. Pricing is value-based and disclosed under NDA to qualified applicants.