On-Prem AI for Pension and Asset Managers: Fiduciary-Grade Intelligence That Stays In-House
Regulated asset and pension managers can now run AI over member and portfolio data inside their own walls, meeting fiduciary, GDPR and conduct duties with a tamper-evident, post-quantum-signed record of every action.
By Micky Irons, founder and CEO of Mickai
The problem fiduciary AI actually has to solve
A pension scheme holds two of the most sensitive datasets in the economy at once: the personal and financial records of its members, and the live positioning of the portfolios that fund their retirement. An asset manager sits on the same combination, plus the trading intent and research edge that defines the firm. Both are bound by duties that do not bend for convenience. The fiduciary standard demands that members' interests come first and that the people running the money can explain their decisions. UK GDPR treats much member data as special-category and gives individuals rights over how it is processed. FCA conduct expectations require that you can show your work.
Generic AI breaks all three at the same point: the moment data leaves the building. Sending member records and portfolio data to a public-cloud model means handing your most regulated assets to infrastructure you do not control, governed by a contract you did not write, exposed to the CLOUD Act and to a chain of subprocessors you cannot fully audit. For a fiduciary, that is not an efficiency trade. It is a breach of the basic promise to keep the beneficiary's interests, and their data, inside trusted hands.
Mickai was built to remove that trade entirely. It is a sovereign AI operating system: AI that a regulated firm owns and runs inside its own walls, on-prem and air-gapped, with every action written to a tamper-evident, post-quantum-signed audit record we call the OAR. The intelligence comes to the data. The data never leaves.
Why on-prem is the only defensible answer
Roughly 850,000 UK businesses, about 15 percent of the economy, and around 5 million across the EU are effectively barred from putting regulated workloads on public-cloud AI. For asset and pension managers the constraints stack: UK GDPR special-category handling, PRA expectations on operational resilience and outsourcing under SS2/21, conduct duties on suitability and record-keeping, and the EU AI Act, which classes credit and insurance-adjacent decisioning as high-risk. Each of those rules points the same way. You must know where the data is, who touched it, what the model did, and why.
Public-cloud AI cannot answer those questions cleanly, because the firm does not own the substrate. On-prem and air-gapped is not a nostalgic preference. It is the architecture that lets a fiduciary keep custody of the data and custody of the reasoning at the same time. Independent sizing puts the sovereign AI market at around 40 billion US dollars in 2025, growing toward roughly 148 billion by 2032. The firms that build for sovereignty early help define that curve rather than chase it.
What the audit record changes
The OAR is the part that turns a model into an instrument a regulator and a trustee can both trust. Every prompt, retrieval, model output and human approval is written to a record that is tamper-evident and signed with post-quantum cryptography, so it stays verifiable for the multi-decade horizon a pension scheme actually lives on. When the Pensions Regulator, the FCA, a scheme auditor or a member's data-subject request arrives, you do not reconstruct what the AI did from memory or logs that can be edited. You produce a signed chain of evidence.
That reframes the AI conversation inside a regulated firm. The question stops being whether you are allowed to use AI and becomes how well you can prove what it did. A signed audit record is the difference between an experiment the risk committee tolerates and a system the board can stand behind.
Built for the asset and pension workflow
Mickai is delivered as a set of Greek-named Studios, each a focused capability that runs against the firm's own data on the firm's own hardware. For this sector the line-up reads directly off the operating model. Plutus covers finance and portfolio analytics. Tyche supports underwriting and risk pricing for the insured and bulk-annuity side. Prometheus drives forecasting across cashflow, funding levels and liability matching. Nemesis runs fraud, AML and surveillance over flows and counterparties. Nomos handles compliance monitoring against your rule set, and Astraea sits alongside for legal. Pythia delivers business intelligence on top, and Aletheia gives you continuous audit. Around the Studios sit Trust Agent, the AMT agentic layer, Vinis voice, OAR-as-a-Service, and HELIOS hardware for firms that want the appliance shipped ready to run.
The point is not a feature list. It is that every one of these runs without a single byte of member or portfolio data crossing the boundary, and every one writes to the same signed record. A scheme can let a model triage member correspondence, model funding scenarios, and screen transactions for AML risk, and still hand its auditor one coherent, verifiable account of everything the AI touched.
The moat under the platform
This is protected work. Mickai LTD holds 104 filed UK patent applications carrying roughly 2,340 claims, with the inventor named as Micky Irons. They are filed, not granted, which matters: it establishes priority and builds a prior-art moat around the sovereign, audited, on-prem approach. Our IP mapping has identified 196 companies and 311 patent-company pairs as potential licensees, including names such as Microsoft, AWS, NVIDIA, Google, Adobe and IBM. That is potential-licensee sizing, not signed revenue, but it shows where the architecture sits relative to the industry: not adjacent to the major platforms, but underneath a problem they will eventually have to solve too.
In June 2026, Crunchbase ranked me at number four globally, with Mickai placing in the top one to two percent of companies on the platform. I read that as a third-party momentum signal and nothing more, but momentum is worth naming when it is independently measured. We are a UK company with Birmingham manufacturing secured, building to scale and heading for the top.
An ally, not a replacement
Mickai is not positioned against the model labs. We are the layer that lets regulated capital adopt advanced AI at all, by giving it sovereignty and a signed audit trail the public cloud cannot. The dual-buyer thesis is straightforward. Regulated firms buy Mickai to deploy AI they could not otherwise touch. The hyperscalers and model providers gain a sovereign on-ramp into the most tightly governed corners of the market. The economics follow the architecture: a year-five revenue path into the billions at high gross margin, underwritten by the IP estate and the breadth of demand. This is a category a hyperscaler would want to own.
The window
The system is built and live. We are now selecting a small number of partner firms in pensions and asset management to deploy against their own data, on their own hardware, ahead of broader availability. This is a moment of choice, not need: the firms that move while the field is still forming will set the standard their peers are later measured against.
If you run, advise or govern a scheme or an asset manager and you want fiduciary-grade AI that never leaves your walls, write to me directly at micky@mickai.co.uk.
FAQ
What is a sovereign AI operating system? It is AI that a regulated firm owns and runs entirely inside its own walls, on-prem and air-gapped, rather than calling out to a public-cloud model. The intelligence runs against the firm's data on the firm's hardware, and the data never leaves the boundary.
How does Mickai help asset and pension managers meet fiduciary and FCA duties? Every prompt, retrieval, model output and human approval is written to the OAR, a tamper-evident, post-quantum-signed audit record. When a regulator, scheme auditor or data-subject request arrives, you produce a signed chain of evidence rather than reconstructing events from editable logs.
Why can these firms not just use public-cloud AI? UK GDPR special-category handling, PRA SS2/21 resilience and outsourcing expectations, FCA conduct duties and the EU AI Act high-risk classification all require custody of the data and the reasoning. Public cloud cannot provide that cleanly because the firm does not own the substrate.
Is Mickai a competitor to the big model labs? No. Mickai is an ally. It is the sovereign layer that lets regulated capital adopt advanced AI at all, and it gives hyperscalers and model providers an on-ramp into the most tightly governed parts of the market.
Is the platform available now? Yes. The system is built and live. We are selecting a small number of partner firms in pensions and asset management to deploy ahead of broader availability.
Frequently asked questions
What is a sovereign AI operating system?
It is AI that a regulated firm owns and runs entirely inside its own walls, on-prem and air-gapped, rather than calling out to a public-cloud model. The intelligence runs against the firm's data on the firm's hardware, and the data never leaves the boundary.
How does Mickai help asset and pension managers meet fiduciary and FCA duties?
Every prompt, retrieval, model output and human approval is written to the OAR, a tamper-evident, post-quantum-signed audit record. When a regulator, scheme auditor or data-subject request arrives, you produce a signed chain of evidence rather than reconstructing events from editable logs.
Why can these firms not just use public-cloud AI?
UK GDPR special-category handling, PRA SS2/21 resilience and outsourcing expectations, FCA conduct duties and the EU AI Act high-risk classification all require custody of the data and the reasoning. Public cloud cannot provide that cleanly because the firm does not own the substrate.
Is Mickai a competitor to the big model labs?
No. Mickai is an ally. It is the sovereign layer that lets regulated capital adopt advanced AI at all, and it gives hyperscalers and model providers an on-ramp into the most tightly governed parts of the market.
Is the platform available now?
Yes. The system is built and live. We are selecting a small number of partner firms in pensions and asset management to deploy ahead of broader availability.