Model Risk (SR 11-7 and PRA SS1/23)
The US Federal Reserve SR 11-7 guidance and the UK PRA SS1/23 supervisory statement set the expectations for model risk management in regulated financial firms: every model must be inventoried, developed and documented soundly, independently validated, governed with clear ownership and remain explainable on demand. Mickai keeps the models, the run logs and the decision lineage on hardware the firm owns, fully offline, so validation and governance are conducted against systems the firm controls end to end rather than a vendor black box. Because every inference is sealed to a post-quantum Open Audit Record with its inputs, features and model version, the firm can evidence exactly how a model reached a figure, which is the core of both frameworks.
Model risk management rests on a demand that is impossible to satisfy with a cloud black box: the firm must be able to inventory every model, evidence how it was developed and validated, govern it with clear ownership, and reproduce exactly how it reached any given output. SR 11-7 sets out the effective challenge, independent validation and ongoing monitoring that a sound model risk framework requires, and PRA SS1/23 codifies equivalent expectations for UK firms across model identification, governance, development, validation and use. None of this is achievable when the weights, the training-data lineage and the run logs sit on infrastructure the firm neither owns nor can audit, and the exposure is compounded when the same shared model serves other customers and drifts outside the firm's control. Mickai keeps the models, the data and the full decision record on hardware the firm owns, so every model is inventoried locally, validated against systems the firm can inspect and reproduce, and monitored continuously through the sealed audit trail. Each inference is logged with the features, the weights version and the reasoning that produced it, so the firm can demonstrate to a supervisor exactly how a model reached a number without a vendor in the loop.
The 6 obligations this framework imposes, each met by construction on hardware you own and mapped to the subsystem that enforces it.
Complete Model Inventory
Mickai maintains a complete inventory of every model in use, its version, its purpose and its owner, on hardware the firm owns, satisfying the SR 11-7 and SS1/23 expectation for a comprehensive model inventory. Because the firm runs its own specialised sovereign models locally, none is hidden inside a vendor tenancy outside the register. Each model and its changes are versioned and sealed to the audit record. The inventory is first-hand, current and reproducible.
Independent Validation Against Owned Systems
Because the models and their data live on hardware the firm owns, independent validation and effective challenge are conducted against systems the firm can inspect, replay and reproduce, rather than a vendor black box. Validation tests and their outcomes are sealed to the Open Audit Record for supervisory evidence. The validators have full access to the weights, inputs and logs the exercise requires. Independent challenge is genuinely possible rather than obstructed by a shared platform.
Decision Lineage and Reproducibility
Every inference is sealed to a causally linked Open Audit Record capturing the inputs, features, model version and reasoning, so the firm can reproduce exactly how a model reached any output. This directly meets the SR 11-7 and SS1/23 expectation that model outcomes be explainable and reconstructable on demand. A supervisor can walk the lineage from a figure back to the originating data. Reproducibility is a property of the architecture rather than a manual reconstruction.
Ongoing Monitoring and Drift Detection
Mickai continuously monitors deployed model behaviour and flags performance drift against the firm's own benchmarks, on hardware the firm controls, meeting the ongoing-monitoring expectation of both frameworks. Because the model does not silently change underneath the firm as a shared cloud model can, drift is attributable to controlled, versioned changes. Monitoring evidence is sealed to the audit record. The firm holds a first-hand, continuous view of model performance.
Model Governance and Clear Ownership
Mickai enforces documented governance, named ownership and change control over every model as signed policy on owned hardware, satisfying the governance expectations of SR 11-7 and SS1/23. Model changes require authorisation and are sealed to the audit record with the responsible owner. There is no vendor updating the model outside the firm's governance. Ownership and control of every model stay firmly with the firm.
Documentation and Explainability Evidence
Mickai captures development, assumption and limitation documentation for each model and seals it to the Open Audit Record alongside the validation evidence, meeting the documentation expectations of both frameworks. Because the documentation is anchored to the actual model version and its sealed run history, it cannot drift out of step with the deployed model. A supervisor receives a coherent, reproducible evidence pack. Documentation is tied to the model it describes rather than maintained separately.
The advantages hold across every framework, and they are architectural, not promotional. The third-party cloud-exposure vector is removed; your own physical, insider and compliance controls remain yours.
The data never leaves your hardware, so no third party and no cloud-provider employee ever sees it. What happens in the server room stays in the server room.
You own the compute and the capability, so the system runs independent of the internet and of any cloud vendor's pricing, terms, or availability.
The data never crosses a geographical or digital border because it never leaves the building, which removes the cross-border-transfer and third-party-processing friction of UK GDPR, Schrems II, and the sector rules. You keep your own obligations.
Fine-tune and run retrieval on your deepest archives to build a hyper-customised co-pilot, with no risk of your proprietary edge training a public model or leaking.
After the hardware and licence, queries cost essentially electricity. A capital asset you own and depreciate, instead of volatile per-token cloud bills.
There is no third-party cloud path, so no competitor and no vendor insider can scrape, intercept, or subpoena your prompts or your fine-tuned weights from the internet. The trust vault is closed by architecture.
You own the software snapshot on your own hardware, so a change to a cloud vendor's terms, a model deprecation, or an outage cannot reach you. The system stays predictable and auditable on-premise as the rules evolve.
How does on-premise AI satisfy SR 11-7 and PRA SS1/23?
Mickai keeps the models, the run logs and the decision lineage on hardware the firm owns, fully offline, so validation and governance are conducted against systems the firm controls end to end rather than a vendor black box. Every model is inventoried locally, independently validated, monitored for drift and governed with clear ownership, and every inference is sealed to a post-quantum Open Audit Record with its inputs, features and model version. The firm can evidence exactly how a model reached a figure, which is the core of both frameworks.
Can we independently validate a model that runs on Mickai?
Yes. Because the models and their data live on hardware the firm owns, independent validation and effective challenge are conducted against systems the firm can inspect, replay and reproduce, with full access to the weights, inputs and logs the exercise requires. Validation tests and their outcomes are sealed to the Open Audit Record for supervisory evidence. Independent challenge is genuinely possible rather than obstructed by a shared platform.
How does Mickai make a model decision reproducible?
Every inference is sealed to a causally linked Open Audit Record capturing the inputs, features, model version and reasoning, so the firm can reproduce exactly how a model reached any output and a supervisor can walk the lineage from a figure back to the originating data. Reproducibility is a property of the architecture rather than a manual reconstruction, which directly meets the explainability expectation of both frameworks.
Does the model change underneath us the way a cloud model can?
No. Because Mickai runs the firm's own specialised sovereign models on owned hardware, there is no vendor silently updating the model outside the firm's governance. Model changes require authorisation and are sealed to the audit record with the responsible owner, so any drift is attributable to controlled, versioned changes rather than a hidden update. Ownership and control of every model stay firmly with the firm.
How is the model inventory kept complete?
Mickai maintains a full inventory of every model, its version, purpose and owner on hardware the firm owns, and because the firm runs its own models locally, none is hidden inside a vendor tenancy outside the register. Each model and its changes are versioned and sealed to the audit record, so the inventory is first-hand, current and reproducible rather than dependent on a vendor's disclosure.
Is Mickai a cloud model service subject to model risk rules elsewhere?
No. Mickai is a Sovereign Intelligence Operating System that runs the firm's own models on hardware the firm owns, acquired as an owned asset rather than a shared cloud service. The public cloud remains useful for non-regulated work; Mickai is the answer for the regulated model-risk boundary where the firm must evidence and control every model itself. It is built on 104 filed UK patent applications covering approximately 2,340 claims, owned by Mickai LTD.
Bring Model Risk (SR 11-7 and PRA SS1/23) in-house.
Briefings are for organisations weighing a sovereign, on-premise deployment. Tell us about your estate and we will walk the obligations, the regulatory crosswalk and the deployment that fits.