Sovereign AI for Universities and Research

A modern research university is a paradox. It exists to publish, to share, to push knowledge into the open, and yet it also guards the most sensitive material in any economy: pre-publication findings, patent-grade discoveries, embargoed clinical data, and the personal records of tens of thousands of students and staff. The moment any of that touches a public inference endpoint, it has left the institution's control, and control is the one thing an academic body cannot afford to lose.

We built Mickai for exactly this boundary. Mickai is a Sovereign Intelligence Operating System, a SIOS that runs on hardware the university already owns, air-gapped or on-premise, with zero data egress. Intelligence comes to the data instead of the data being shipped to a cloud, and every result it produces arrives with a signed, verifiable record of how it was made. For a sector whose entire currency is provenance and trust, that changes what artificial intelligence is allowed to touch.

The data a university cannot afford to leak

Consider what flows through a single faculty of engineering in a week. A doctoral student uploads three years of unpublished experimental results to summarise them for a thesis chapter. A principal investigator runs a draft grant application, worth several million pounds, through a model to tighten the argument. An administrator batch-processes the mental-health disclosures of first-year students to flag those at risk. Each of these is a legitimate use of intelligence. Each, on a public endpoint, is a data-protection incident waiting to be named.

The regulatory weight here is real. The General Data Protection Regulation (GDPR) governs every student record, and the European Union Artificial Intelligence Act (EU AI Act) now classes much educational and research profiling as high-risk. Where clinical or biomedical research is involved, the Health Insurance Portability and Accountability Act (HIPAA) and its equivalents apply, and dual-use research can fall under the International Traffic in Arms Regulations (ITAR). A university does not get to plead that a third-party model provider was the one who saw the data. Under the ISO 42001 standard for artificial intelligence management systems and the National Institute of Standards and Technology Artificial Intelligence Risk Management Framework (NIST AI RMF), the accountability stays with the institution.

Intelligence that never leaves the building

Mickai inverts the usual arrangement. Rather than sending prompts and documents to a remote data centre, the SIOS installs on the university's own machines, from a single secure workstation in a research group to a rack in the central data centre. The brains that do the reasoning, revocable and individually governed, run locally. Sensitive corpora, the unpublished papers, the genomic sets, the student information systems, are indexed in place and never copied out.

This is not a private cloud with a comforting label. It is a genuine layer that sits alongside the operating systems the institution already trusts, dedicated to artificial intelligence and bounded so that data cannot egress. The public cloud giants remain allies for the workloads that belong in the open, and Mickai deliberately serves the regulated boundary those providers cannot cross on the customer's own terms. When a virologist reasons over an embargoed dataset, or a legal office reviews a confidential technology-transfer agreement, the work happens inside the university's own walls and stops there.

Signed provenance on every result

Reproducibility is the currency of science, and the reproducibility crisis has already taught the sector how fragile a result is when its origins are unclear. Generative intelligence, left ungoverned, makes this worse: an answer with no record of the model, the data, and the reasoning behind it is worse than no answer at all, because it wears the costume of authority.

Mickai attaches an Operation Attestation Record (OAR) to every action the system takes, and it does so before the action executes rather than after. The OAR captures which brain ran, against which version of which dataset, under whose authority, and with what result. Each record is signed with post-quantum signatures, specifically Federal Information Processing Standard 204 Module-Lattice Digital Signature Algorithm (FIPS 204 ML-DSA-65), and linked into a SHA-3-512 hash-linked chain. The outcome is a tamper-evident, cryptographically-signed audit ledger of the institution's entire research pipeline.

For a peer reviewer, a funding body, or a patent examiner, this is transformative. A finding produced through Mickai can be verified offline, without contacting any vendor, showing precisely how it was derived. Provenance stops being a promise and becomes mathematics.

Protecting the intellectual property that pays for the campus

University technology-transfer offices turn research into licensing income, spin-outs, and patents, and that value depends entirely on discoveries staying confidential until they are properly protected. A single pre-publication leak can vaporise the novelty a patent requires. Feeding a promising result into a public model to help draft a claim is, in effect, a public disclosure, and it can extinguish the very rights the office exists to secure. The capability matters here for us too: our own 104 filed UK patent applications, about 2,340 claims owned by Mickai LTD, exist precisely because the discoveries behind them were kept sovereign until they were protected.

Because Mickai runs on owned infrastructure with no data egress, researchers can reason over their most valuable findings without ever crossing that line. High-stakes actions, releasing a dataset to a collaborating institution, exporting a model, approving a disclosure, can be gated behind multi-brain plus voice-biometric approval, so no single account and no single compromised credential can move sensitive intellectual property out of the institution. We built the SIOS on the principle that the researcher, not the platform, decides what the world gets to see and when.

Governing thousands of users without surveilling them

A university is not a corporation with a tidy hierarchy. It is a federation of faculties, institutes, and independent research groups, each fiercely protective of its own autonomy. Any intelligence system imposed across it must respect that structure rather than flatten it. Mickai's brains are individually revocable and scoped, so a clinical-research group can grant one set of capabilities while the humanities faculty grants another, and a single institution can host many sovereign boundaries under one roof.

Crucially, governance here is not surveillance. The audit ledger records what the intelligence did, not what people privately thought or typed. Because everything runs locally and offline, the institution proves compliance with GDPR and the EU AI Act to a regulator without a third party ever observing its members. Accountability and privacy, so often traded off against each other, hold together at the same time.

The bottom line

Universities and research institutes carry a duty that few other organisations face: to be radically open with knowledge and radically protective of the data and discoveries that produce it. Public inference endpoints force a choice between those duties. A Sovereign Intelligence Operating System dissolves the choice.

With Mickai, intelligence runs on infrastructure the university owns, sensitive data never leaves the building, and every result carries a signed, offline-verifiable record of its own making. Research IP stays confidential until it is protected, student data stays private by architecture rather than by policy, and provenance becomes something a reviewer can check with cryptography instead of trust. For a sector built on both openness and integrity, that is not a compromise. It is the standard the work has always deserved.