Sovereign AI for Capital Markets: Research, Surveillance and MNPI Behind the Wall

By Micky Irons, founder and CEO of Mickai

The data a trading floor cannot send anywhere

Capital markets run on information that is dangerous by design. Draft research before publication. The pre-announcement deal book. Order flow that reveals client intent. Surveillance feeds built specifically to catch insider dealing and market abuse. Each of these is, in regulatory terms, the kind of material non-public information (MNPI) that a firm is legally obliged to control, restrict and ring-fence.

Public-cloud AI asks a buy-side or sell-side firm to do the one thing it cannot do: take that material, send it outside the wall, and let a third party process it on infrastructure the firm does not own and cannot audit at the silicon level. That is not a workflow problem. It is a regulatory and fiduciary one.

This is the gap Mickai was built to close. Mickai is a sovereign AI operating system, an SIOS: AI that a regulated business 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. Built and live. The premise is simple. If the model runs where the data already legally lives, the data never has to leave.

Why the desk cannot use public-cloud AI

The barriers here are not cultural caution. They are specific obligations that a compliance officer cannot wave away.

Information barriers are the foundation of the equities and advisory business. A firm maintains Chinese walls precisely so that research, corporate finance and trading cannot see each other's MNPI. Pushing all of that into a shared external model collapses the separation the regulator requires the firm to prove it maintains.

Market-abuse surveillance under MAR and the wider supervisory regime depends on a firm monitoring its own communications and trading for insider dealing and manipulation. That surveillance corpus is among the most sensitive data a firm holds, because it concatenates everything: chats, voice, orders, news. Outsourcing the analysis of it to a public model means exporting the firm's entire behavioural picture.

Operational resilience expectations, including PRA SS2/21 on outsourcing and third-party risk, push firms to understand, control and exit their dependencies. A black-box AI dependency that processes MNPI is exactly the kind of concentration risk supervisors are now asking firms to map and govern.

Cross-border legal exposure compounds it. The CLOUD Act means data on a US hyperscaler can be compelled by US authorities regardless of where the servers sit. For a firm holding the order flow and deal books of clients across jurisdictions, that is not a theoretical footnote. It is a reason counsel says no.

The result is a structural barrier. By our sizing, roughly 850,000 UK businesses, about 15 percent, and around 5 million across the EU sit under rules that effectively bar them from sending their most valuable data to public-cloud AI. Capital markets firms are squarely inside that population.

What changes when the stack is owned

Mickai inverts the model. Instead of sending data to the intelligence, the intelligence runs where the data already sits, behind the firm's own controls, air-gapped where required.

For a markets business that reshapes three jobs at once.

Research. Analysts can run drafting, summarisation, document comparison and thematic synthesis across the full research library and the firm's own primary data without a single token crossing the perimeter. Embargoed notes stay embargoed because they never leave the building.

Surveillance. The same engine reads the firm's communications and trading data to flag potential insider dealing, spoofing and manipulation, and writes every detection and every analyst review into the OAR. When the regulator asks how a conclusion was reached, the firm produces a signed, tamper-evident record rather than a screenshot.

MNPI handling end to end. Because the model is inside the wall, the deal book, the order book and the surveillance corpus can all be processed under the firm's existing information-barrier policy. The AI becomes another controlled system on the inside, not an exfiltration route to the outside.

We deliver this through a set of Greek-named Studios, each a focused capability that sits on the same sovereign substrate and the same audit record. For markets that means Nemesis for fraud and AML, Plutus for finance, Nomos for compliance, Astraea for legal, Aletheia for audit and Pythia for business intelligence, alongside Trust Agent, the AMT agentic marketing system, Vinis voice, OAR-as-a-Service and the HELIOS hardware line. One substrate, one audit spine, many desks.

The moat under the product

Sovereignty as a category needs defensibility, and that is where the IP estate sits. Mickai LTD holds 104 filed UK patent applications, roughly 2,340 claims, inventor Micky Irons. Filed, not granted, and we say so plainly. What filing secures is priority and a prior-art position: a moat built on the architecture of running regulated AI inside the wall with a post-quantum-signed audit trail.

That estate is broad enough that our own mapping identifies 196 companies and 311 patent-company pairs as potential licensees, a list that includes Microsoft, AWS, NVIDIA, Google, Adobe and IBM. That is potential-licensee sizing, not signed revenue, and it should be read as such. It points to the same conclusion from a different angle: the sovereign layer is something the largest infrastructure players have a reason to want access to.

Mickai is an ally to that ecosystem, not an OpenAI-killer. The dual-buyer thesis is the point. Regulated enterprises buy the owned stack because they have no compliant alternative. The platforms have a reason to license the architecture because sovereignty is the layer their regulated customers keep asking for and cannot currently get. Both buyers are served by the same patented substrate.

The market and the momentum

The sovereign AI market is sized at around USD 40 billion in 2025, on a path to roughly USD 148 billion by 2032. Capital markets is one of its densest pockets, because the data is both the most valuable and the most legally constrained in any sector.

As a dated third-party signal: in June 2026, Micky Irons was ranked number four on Crunchbase, with the Mickai company profile placing in the top one to two percent globally. We read that as momentum, not arrival. Mickai is a UK company with Birmingham manufacturing secured, building to scale.

The economics follow the structure. A Year-Five revenue path to billions at high gross margin is underwritten by the IP estate and the dual-buyer thesis. It is the kind of sovereign layer a serious infrastructure player would rather own than compete with.

Where this is heading

The build is live and the path is set. As regulated AI moves from optional to mandatory inside capital markets, the firms that own their stack will be the ones that can use AI on their most valuable data at all. The owned, on-prem, audit-spined model is not a feature of that future. It is the precondition for it, and it is the layer the category is now organising around.

If sovereign AI for research, surveillance and MNPI is a question on your desk or in your portfolio, you can reach me directly at micky@mickai.co.uk.

FAQ

Why can't capital markets firms just use public-cloud AI with a private tenant? A private tenant still processes MNPI on infrastructure the firm does not own and cannot audit at the silicon level, and it remains exposed to compulsion under laws like the CLOUD Act. Information barriers, MAR surveillance obligations and outsourcing expectations such as PRA SS2/21 push firms toward control they can prove. Mickai keeps the model inside the firm's own walls so the data never leaves.

What does Mickai actually do for a trading floor? It runs research drafting and synthesis, market-abuse surveillance and end-to-end MNPI handling on an owned, on-prem and where required air-gapped stack, with every action written to a tamper-evident, post-quantum-signed audit record, the OAR.

Is Mickai a competitor to the major AI platforms? No. Mickai is an ally. The dual-buyer thesis means regulated firms buy the owned stack while the large platforms have a reason to license the sovereign architecture their regulated customers are asking for.

Are the 104 patents granted? They are filed UK applications, roughly 2,340 claims, held by Mickai LTD. Filing secures priority and a prior-art moat. We state the filed-not-granted position plainly.

Frequently asked questions

Why can't capital markets firms just use public-cloud AI with a private tenant?

A private tenant still processes MNPI on infrastructure the firm does not own and cannot audit at the silicon level, and it remains exposed to compulsion under laws like the CLOUD Act. Information barriers, MAR surveillance obligations and outsourcing expectations such as PRA SS2/21 push firms toward control they can prove. Mickai keeps the model inside the firm's own walls so the data never leaves.

What does Mickai actually do for a trading floor?

It runs research drafting and synthesis, market-abuse surveillance and end-to-end MNPI handling on an owned, on-prem and where required air-gapped stack, with every action written to a tamper-evident, post-quantum-signed audit record, the OAR.

Is Mickai a competitor to the major AI platforms?

No. Mickai is an ally. The dual-buyer thesis means regulated firms buy the owned stack while the large platforms have a reason to license the sovereign architecture their regulated customers are asking for.

Are the 104 patents granted?

They are filed UK applications, roughly 2,340 claims, held by Mickai LTD. Filing secures priority and a prior-art moat. We state the filed-not-granted position plainly.