On-Premise Fraud Detection AI: Anti-Money-Laundering That Stays Air-Gapped

On-premise fraud detection AI is financial-crime surveillance that runs on the institution's own hardware, scoring un-redacted transactions for fraud and money-laundering risk without a single record leaving the building. Anti-money-laundering work demands the model see the real, complete transaction picture, the very data that is most dangerous to send anywhere. The Mickai Sovereign Intelligence Operating System (SIOS) brings the compute to the data, behind the firewall, so what happens in the server room stays in the server room. It is the sovereign, air-gapped alternative to cloud financial-crime platforms such as SAS AML, NICE Actimize and Featurespace.

What on-premise fraud detection AI does

Financial-crime surveillance is pattern recognition at scale: spotting the structuring, the unusual counterparty, the velocity spike, the anomaly that does not fit the customer's known behaviour. Done well, it needs the un-redacted record, the full transaction, the real names, the actual amounts, the genuine network of relationships, because the signal lives in exactly the detail that anonymisation strips away.

The Mickai fraud studio, **Nemesis**, performs this work entirely inside the institution's perimeter:

• Anomaly and outlier detection across transaction, payment and account-activity streams.
• Money-laundering typology surveillance, scoring structuring, layering and unusual-counterparty patterns.
• Network and relationship analysis to surface the connections a single-transaction view would miss.
• Alert triage and prioritisation, so investigators spend their time on the cases that matter.

Nemesis runs as one of Mickai's own sovereign brains, fine-tuned on the institution's own history so it learns the institution's genuine baseline, with the Mickai sovereign vector store holding every embedding locally. The fine-tuning is local and stays local.

What it replaces

For decades, financial-crime surveillance has meant a small set of established platforms: SAS AML, NICE Actimize and the machine-learning fraud engines such as Featurespace. These are capable systems. The difficulty is where the most modern AI versions of them increasingly want the data to live, and what the institution gives up when un-redacted financial-crime data is processed outside its own perimeter.

“The cloud is the exfiltration vulnerability vector. Anti-money-laundering surveillance on un-redacted transaction data belongs on hardware the bank owns, not on a shared platform the bank cannot inspect.”

Mickai is the air-gapped alternative. It delivers the surveillance capability while keeping the data, the model and the audit trail behind the institution's own firewall, with no internet path to the records.

Why financial-crime data cannot go to the cloud

The transaction record is the most sensitive material a financial institution holds, governed by financial secrecy, data-protection law and the cross-border transfer question that arises the moment data moves jurisdiction. Pushing un-redacted transaction data to a cloud model is third-party processing, with a cross-border transfer risk layered on top.

A Data Processing Agreement does not undo this. Once the record leaves the institution, it sits on infrastructure the institution does not own, governed by terms that can drift, exposed to an attack surface the institution cannot audit. A signed contract does not stop a breach at the provider, an outage that takes surveillance offline, or interception in transit. The protected data has still left the building. For a financial-crime function whose entire purpose is to control risk, building surveillance on that exposure is a contradiction.

There is also the permanence problem. An institution that builds its surveillance on a cloud vendor inherits that vendor's policy drift, price volatility and terms-of-service changes. Mickai removes the dependency through model and weight ownership: the institution owns the snapshot, insulated from a vendor changing the rules and from the EU AI Act reshaping the ground under a rented service.

Why on-premise wins: the architecture

Mickai deploys a single-tenant operating system inside the institution, on hardware the institution owns, with deterministic network isolation and locally contained inference. The design principle is Compute-to-Data: bring the model to the records, not the records to the model. The benefits for a fraud and financial-crime function are direct:

• **Zero data egress.** The transaction record is scored entirely inside the perimeter. The architecture removes the cross-border transfer and third-party processing path, so data residency holds because the data never moves.
• **Un-redacted analysis.** The model can finally see the full, real record rather than a sanitised subset, because the data never leaves a place it is allowed to be. That is where the signal lives.
• **Reduced attack surface.** With no internet path to the records, the attack surface is reduced to what the institution itself controls. The honest boundary: this reduces, it does not remove, risk, insider and physical-access exposure remain the institution's to manage.
• **Predictable economics.** The compute is a fixed, depreciable capital asset rather than a volatile per-token cloud bill. Above a meaningful volume of surveillance, the marginal cost of another query approaches zero. This is CapEx AI optimisation, not OpEx exposure.
• **Resilience.** The system runs independent of cloud outages because the institution owns the compute.

Who needs it

This is built for the institutions where financial-crime surveillance is both mission-critical and legally constrained: private and global banks, where Nemesis sits alongside the wider finance, compliance and underwriting studios; insurers running claims-fraud surveillance; payment firms and any regulated institution that must monitor money movement on data it cannot expose. It is a natural anchor for a private-banking deployment and for any sector where un-redacted financial data is the working material.

The honest framing throughout: this removes a major category of exposure and gives the institution true data sovereignty over its surveillance, but the institution keeps its own regulatory obligations, its own model-governance duties, and its own internal controls. Mickai gives those obligations a sovereign place to run.

What makes Mickai different

Mickai is built and owned, not rented. Three properties set it apart from a cloud platform with an on-premise badge:

1. **The Open Audit Record (OAR).** Every alert, score and material action is written to a signed, inspectable record. When a regulator or an internal auditor asks why the system flagged or cleared a transaction, the answer is a verifiable artefact, not a vendor's assurance. For a function that lives or dies on its audit trail, governance engineered into the system is the point. 2. **A defensible moat of 101 filed UK patent applications.** The sovereign architecture is protected by 101 filed UK patent applications owned by Mickai LTD, covering the Compute-to-Data design, the audit record and the hardware-bound identity model. This is durable, original intellectual property, not a wrapper around someone else's engine. 3. **Hardware-bound identity.** The deployment's identity is tied to the physical machines it runs on, so the surveillance system and the data it reads cannot be silently cloned or moved off the institution's iron.

Mickai was built by Micky Irons, founder, chief executive and named inventor, who designed the SIOS for exactly the functions, financial-crime surveillance among them, whose data is too sensitive and too regulated to send to a shared cloud.

Request a private demonstration

If you are a Chief Operating Officer, Chief Information Officer, Chief Information Security Officer, Chief Financial Officer, money-laundering reporting officer or General Counsel, request a private demonstration of on-premise fraud detection AI on the Mickai SIOS. We will show anomaly scoring and anti-money-laundering surveillance running on un-redacted transaction data entirely on hardware you own, with the Open Audit Record open for inspection and not a single record leaving the building.