Every Credit Decision, Sealed and Explainable
Consumer Duty asks finance and retail credit to prove every consequential decision was fair, auditable, and explainable, and a cloud trail the customer cannot verify does not clear that bar.
A decision becomes a record, whether you keep one or not
A consumer applies for store credit at the till. A buy-now-pay-later split is offered on a kitchen appliance. An overdraft extension is requested inside a banking app. In the seconds that follow, a model weighs affordability, fraud signals, prior history, and a risk threshold, and a system says yes or no. Under the FCA's Consumer Duty, that moment is no longer a private calculation. It is a consequential decision, and the firm must later be able to show that it was fair, that the customer could understand it, and that the basis for it can be reconstructed.
The hard part is what "reconstructed" means when the model lives in a shared cloud. A firm can produce a log. It can show a timestamp, an input vector, a score, an outcome. What it cannot do, from outside a multi-tenant system it does not control, is prove that the log is the truth. It cannot prove the model version that ran was the one in its model-risk register. It cannot prove the record was not regenerated, reformatted, or quietly aligned to the answer the firm wishes it had given. The audit trail becomes an assertion about the past rather than evidence of it.
This is the gap the Mickai Sovereign Intelligence Operating System closes for consumer-facing finance and retail credit. The decision still happens in milliseconds. The difference is that it leaves behind something a regulator, an ombudsman, or a court can verify without taking the firm's word for it.
What Consumer Duty actually demands of a decision
The Consumer Duty, in force since 2023, is not a disclosure rule in new clothes. It asks firms to deliver good outcomes and to evidence that they have. For any consequential decision, three things have to hold together at once. The decision must be explainable in terms the customer can understand. It must be auditable, meaning the firm can return to it and show how it was reached. And it must survive the cross-examination of hindsight, when a complaint lands eighteen months later and the question is not "what does your system say now" but "what did your system do then".
Layer the rest of the regulatory perimeter on top and the burden compounds. UK GDPR Article 22 constrains solely automated decisions with legal or similarly significant effect and gives the customer a right to meaningful information about the logic involved. The Consumer Credit Act and the Consumer Rights Act sit underneath the lending itself. For any firm touching cards or payments, PCI-DSS governs the data the model is reading. And SYSC, the FCA's systems-and-controls regime, expects model governance to be real rather than aspirational.
None of these is satisfied by "the cloud provider keeps logs". They are satisfied by a record the firm holds, controls, and can prove was not altered.
Why a cloud trail the customer cannot verify falls short
There is a category error at the heart of relying on a vendor's log. The firm is the accountable party. The FCA does not pursue the cloud provider when a credit decision is found to have been unfair. It pursues the firm. Yet in a shared cloud the firm is asking a third party to vouch for the evidence that will be used against the firm. The provider controls the storage, the retention, the access, and the format. The firm controls none of it and can verify none of it from the outside.
“If you are a multibillion-dollar company running on Anthropic or OpenAI, and your direct competitor of comparable scale sits on the same vendor stack, what stops them paying a vendor insider to leak your data, your tactics, your leads, your sales strategy? Inside a third-party cloud, there is no safeguard you can verify from the outside. The only answer is a sovereign system where you hold the keys, with no third-party cloud data path.”
That is the architectural truth, stated by Micky Irons, founder and CEO, Mickai LTD. An internal log a firm cannot independently prove is intact is not the same as evidence. It is a claim about evidence. When the stakes are an ombudsman ruling or a skilled-persons review under SYSC, the distinction is the whole case.
How Mickai seals the decision
In the Mickai SIOS, consumer-credit and fraud decisioning runs through Nemesis, the fraud and anomaly studio, on hardware the firm owns. The customer's data never leaves the building. There is no third-party data path to a shared model. And every consequential decision Nemesis touches is sealed to the Open Audit Record, the OAR, at the moment it is made.
The OAR is the difference between a log and proof. Each decision is bound to a post-quantum signature that captures what mattered. The inputs the model saw. The model and inference substrate that ran, by verifiable provenance rather than by label. The score, the threshold, the outcome, and the time. The seal is produced inside the firm's own environment, under keys the firm holds. Anyone with the public verification material can later confirm that the record is the original and has not been edited. The firm does not ask the regulator to trust it. It hands the regulator something the regulator can check.
This is what turns a defensible-sounding process into a defensible one. When the complaint arrives, the firm does not reconstruct what probably happened. It produces the sealed record of what did happen, and the seal proves it.
Model-risk discipline, the way the supervisors expect
Sealing a single decision is necessary but not sufficient. The regulators that matter to lenders, the PRA in the UK and the supervisory model behind SR 11-7 in the US, expect the model itself to be governed across its whole life. SR 11-7 is the canonical text on model risk: development, validation, ongoing monitoring, version control, and an inventory in which every model in production is accounted for.
A shared cloud frustrates this at the root, because the firm cannot pin the substrate. The model behind an API can change without the firm's knowledge or sign-off. The version in the model-risk register becomes a hopeful description rather than a controlled fact. Inside Mickai, the inference substrate is sealed and sits in the firm's own model-risk register with verifiable provenance. The version that decided a customer's application is the version the OAR records, and the version the register lists, and the two can be proven identical. Validation, monitoring, and challenge all run against a substrate the firm controls rather than one it rents and cannot inspect.
There is a quieter benefit here too. Cloud models drift.
“When companies use the Mickai Sovereign Intelligence Operating System, the context-compression problem that plagues cloud LLMs is removed at the architectural level. Cloud systems hallucinate and drift off topic because shared multi-tenant storage forces aggressive context compression, summary-pass swaps, and lossy recall. Inside Mickai, the operator owns the memory. They expand it inside their own data centre or workstation, scale it on Poseidon rack-scale or local NVMe, and never compete with another tenant for context budget. The result is a measurable reduction in drift and hallucination.”
For a decisioning system that must behave consistently across thousands of borderline cases, that consistency is not a nicety. It is the ground on which the firm stands behind its outcomes.
The retail credit reality
Bring this down to where it bites. A major electronics and appliance retailer offering point-of-sale finance. An employee-owned hi-fi and home-cinema chain with an FCA credit-broking core. Each holds identity, purchase history, payment and card data, and consumer-credit data, and each must make affordability and fraud decisions that fall squarely inside Consumer Duty. Running those decisions through a shared cloud AI puts regulated, card-bearing, identity-rich data on a path the firm cannot audit, and produces decisions the firm cannot prove.
The Mickai retail vertical pack answers this end to end. Nemesis seals every consumer-credit and fraud decision to the OAR for Consumer Duty. Nomos, the compliance studio, produces the DPIA, the PCI map, and the signed compliance artefact that turns "we cannot use AI here" into "we can, and here is the proof". Iris handles support with PII that never leaves the building, and Xenia personalises on owned data with no customer record sent to the cloud. The decision is fast, the data is sovereign, and the record is sealed.
The close
Consumer Duty did not ask firms to make better decisions in private. It asked them to make decisions they can prove were fair, in public, on demand, long after the moment has passed. A log a firm cannot verify does not meet that test, because it is a story about evidence rather than the evidence itself. The firms that hold up under the next skilled-persons review will be the ones that can hand over a sealed, verifiable record of exactly what their model did and why. Mickai builds that record into the decision itself. Hold your keys, keep the data in the building, and let the seal do the arguing.