Use case · AI / regulation / control plane
Revolut × AI-Regulation x Sovereign/Local Models x FOL x Google Cloud
The interesting problem is not “explainable AI”.
It is whether a licensed digital bank can govern an automated decision while it is happening: which rule applies, who owns the override, what evidence survives, and what changes when the model, the product, or the law moves.
This note compresses three public artifacts - during my Sept 2026 week sprint - into one claim. It is an outsider working note for the unofficial&personal book project revolut.hot. Speculation is labelled.
Revolut x AI-Regulation x Sovereign/Local Models x FOL
- Global AI Regulation Changes 2026-2030 — jurisdiction map, timeline, operational surface.
- FOL / verifiable controls — where first-order-logic helps (and where it does not).
- Let's add NVIDIA — Nvidia's Nemotron x FOL x compliance.
- Let's add Google Cloud — local AI decisions with cloud additional witness.
- Let's add Human & Reality — real-life compliance experience edits and add-ons.
Call that capability algorithmic sovereignty: not autarky, not “run every model on-prem”, and not a proof that the institution is legally compliant. It is the practical ability to constrain execution and show the trace. Frontier models can still draft, retrieve, and reason. They should not be the last authority on a credit, fraud, or financial-crime action.
What the public record already forces:
The EU path is still a horizontal risk statute (navigator: Reg. 2024/1689, plus the later Omnibus timetable). The UK path is still sectoral conduct: Consumer Duty, SM&CR, PRA model-risk language. Those are different machines. A bank that treats “AI policy” as one memo will fail one of them.
Not every banking model is Annex III high-risk. Fraud systems sit in a different pocket from creditworthiness tools. Nightly retraining can still become a “substantial modification” problem if the same weights leak into a high-risk pipeline. The United Kingdom has not passed a general AI Act for financial services. These corrections are in the DOI-preprint; they matter more than the neon.
A control plane, not a religion
- 01 perception
Models propose: score, draft, retrieve, cluster, write a first brief. - 02 policy
A deterministic layer evaluates structured facts against a versioned rule. SAT / fail-closed / escalate. - 03 evidence
Inputs, model version, policy version, reason codes, human override, outcome. One packet, not a slide.
Datalog, policy-as-code, SMT, and proof assistants belong in layer 02 and in offline audit of layer 03. They do not belong on a 50 ms fraud path unless the rule is a cheap threshold. They do not translate “reasonable steps” into a theorem. FOL-page verdict: useful, not sufficient.
Not another stack diagram. A measurable split of work:
- Tasks where a cheaper model is already enough (boilerplate, first draft, routing).
- Tasks where quality is judged on a large held-out set — tens or hundreds of thousands of cases — not a demo.
- Tasks where the scarce resource is not tokens but hours to “safe to ship”: review, reason codes, incident path.
If open-weight models keep losing on the large-set comparison, that is not an argument against a policy layer. It is an argument that the policy layer should sit on top of whatever model currently wins, including a frontier API and an in-house fine-tune.
Sources in this bundle
- The NVIDIA Innovator's Dilemma →
- Europe's First Trillion-Dollar Startup? →
- FuturoShock.AI →
- + with dash of compliance-background:)
- New: Algorithmic Sovereignty in Digital Banking — obligations → runtime controls.