Tyche: On-Prem Underwriting Intelligence That Keeps Policyholder Data Sovereign
Tyche brings AI-assisted risk scoring and pricing to insurers while special-category and health data stays air-gapped, on-prem, and fully auditable.
The problem underwriters live with every day
Underwriting is a data-hungry discipline. To price risk well, an insurer reasons over medical histories, claims records, financial profiles, occupational data, and a long tail of special-category information that sits at the most sensitive end of UK GDPR. The better the model, the more of that data it wants to see.
That is exactly where public-cloud AI breaks down for a regulated insurer. Sending health records and special-category data to a third-party model endpoint is not a procurement question, it is a lawfulness question. The PRA expects firms to understand and control their models under SS2/21. UK GDPR places special-category and health data behind stricter conditions. The US CLOUD Act means data held by a US-headquartered provider can be reachable regardless of where the servers physically sit. For many carriers, the honest answer to "can we put policyholder health data into that cloud model" is simply no.
So underwriting teams are stuck. They can see what AI-assisted risk scoring would do for loss ratios, speed-to-quote, and pricing accuracy, but they cannot lawfully feed the model the data that would make it useful. Tyche exists to close that gap.
What Tyche is
Tyche is the underwriting Studio inside Mickai, the sovereign AI operating system. Mickai is AI that regulated businesses own and run inside their own walls, on-prem and air-gapped, with every action written to a tamper-evident, post-quantum-signed audit record we call the OAR. This is built and live, not a concept.
Tyche runs risk scoring, pricing support, and case triage where the data already lives. The model comes to the data. The data never leaves the building. There is no outbound call to a public-cloud endpoint, because the intelligence is installed inside your own environment, behind your own firewall, on hardware you control. For special-category and health data, air-gapped is not a nice-to-have. It is the only configuration that lets you use the data at all.
How it works in an underwriting workflow
A submission arrives. Tyche reads the structured fields and the messy unstructured material that usually slows a human underwriter down: doctor's letters, claims narratives, schedules, supporting documents. It surfaces the risk factors, proposes a risk score with the reasoning attached, and flags the cases that need a senior human decision rather than a fast-track quote.
Three things make that safe inside a regulated carrier.
First, every step is written to the OAR. When Tyche reads a field, weighs a factor, or proposes a score, that action is recorded in a tamper-evident, post-quantum-signed log. If a regulator, an internal model-risk team, or a complaints process later asks why a particular applicant was priced the way they were, the answer is an auditable record rather than a shrug.
Second, the model is explainable by construction. Tyche presents the factors behind a score, not just the number. Underwriters stay in control and can overrule, and the override is logged too. This matters for treating customers fairly and for demonstrating that pricing decisions are governed, not opaque.
Third, nothing leaves. Because Tyche runs on-prem and air-gapped, the special-category data that makes underwriting accurate stays inside your own boundary throughout. You get the lift without the lawfulness problem.
Why sovereign, and why now
The wedge here is not theoretical. Roughly 0.85 million UK businesses, about 15 percent, are in the band that legally cannot push their data to public-cloud AI. Across the EU that figure is closer to 5 million. The drivers are stacking up at once: PRA SS2/21, UK GDPR special-category conditions, the EU AI Act treating much of insurance pricing as high-risk, the NIS Regulations, and the reach of the US CLOUD Act over US-headquartered providers. The sovereign AI market reflects this, growing from around USD 40 billion in 2025 toward an estimated USD 148 billion by 2032.
Mickai is built for that band specifically. We are not asking insurers to trust a cloud they are not allowed to use. We are giving them the AI to own and run themselves.
As a dated third-party momentum signal, Micky Irons was ranked number 4 on Crunchbase by CB Rank (Person), verified live in June 2026, with the Mickai company profile in the top one to two percent globally. That is a snapshot of attention at a point in time, not a permanent claim, and we read it as a sign we are building toward something the market wants.
Where Tyche sits in the wider platform
Tyche does not work alone. It is one of a family of Greek-named Studio modules inside Mickai: Nemesis for fraud and AML, Plutus for finance and FP&A, Prometheus for forecasting, Iris for customer service, Nomos for compliance, Astraea for legal, Panacea for clinical, Pythia for business intelligence, and Aletheia for audit, alongside Trust Agent, AMT, Vinis voice, and OAR-as-a-Service. For an insurer, that means underwriting intelligence, fraud detection, and compliance oversight can share one sovereign substrate and one audit record, rather than living in disconnected tools that each touch sensitive data on their own terms.
The intellectual property underneath is substantial. Mickai LTD holds 104 filed UK patent applications, roughly 2,340 claims, with Micky Irons as inventor. These are filed, not granted, and they exist to establish priority and a prior-art moat around how sovereign, auditable, on-prem AI is built. Separate analysis identifies 196 companies across 311 patent-company pairs as potential licensees, including names like Microsoft, AWS, NVIDIA, Google, Adobe, and IBM. To be precise, that is potential-licensee sizing, not a claim that anyone is infringing.
The dual-buyer point
A quick word on positioning, because it matters. Mickai is an ally, not an OpenAI killer. The public-cloud labs are extraordinary at building frontier models. What they cannot offer a regulated insurer is data that never leaves the building. That is the gap Mickai fills. Some buyers will run frontier cloud models for everything that is allowed to leave, and Mickai for everything that is not. Both can be true in the same carrier.
A window for selected partners
Mickai is a UK company, with Birmingham manufacturing secured, and we are building to scale. As part of that, a pre-seed window is open to a small number of selected partners. This is an opportunity to get involved early with a platform that is already built and live, never a pitch born of desperation. If you are an insurer, a reinsurer, or an investor who understands why sovereign underwriting intelligence is now a requirement rather than a luxury, I would like to talk.
Reach me directly at micky@mickai.co.uk.
Micky Irons, founder and CEO of Mickai.
Frequently asked questions
Can insurers legally use AI on special-category and health data with Tyche?
Yes. Tyche runs on-prem and air-gapped inside the insurer's own environment, so special-category and health data never leaves the building. There is no outbound call to a public-cloud model endpoint, which removes the lawfulness problem that blocks cloud AI under UK GDPR special-category conditions and the reach of the US CLOUD Act.
How does Tyche keep underwriting decisions auditable?
Every action Tyche takes, from reading a field to weighing a factor to proposing a score, is written to the OAR, a tamper-evident, post-quantum-signed audit record. If a regulator, a model-risk team, or a complaints process asks why an applicant was priced a certain way, the answer is an auditable record. Underwriter overrides are logged too.
Does Tyche replace underwriters?
No. Tyche surfaces risk factors and proposes a score with its reasoning attached, then flags the cases that need a senior human decision. Underwriters stay in control and can overrule it, which keeps pricing governed and supports treating customers fairly.
Is Mickai trying to replace public-cloud AI labs?
No. Mickai is an ally, not an OpenAI killer. Frontier cloud labs are excellent at building models. What they cannot offer a regulated insurer is data that never leaves the building. Many carriers will run cloud models for what is allowed to leave and Mickai for what is not.
Is Mickai a concept or a working platform?
Built and live. Mickai is a UK company with Birmingham manufacturing secured, holding 104 filed UK patent applications with roughly 2,340 claims. A pre-seed window is open to a small number of selected partners as the company builds to scale.