Sovereign AI for Higher Education and Research: Protecting Export-Controlled and Funded Research

By Micky Irons, founder and CEO of Mickai

The research the cloud cannot hold

A modern university is one of the most complex data-control environments in existence. Inside a single campus you find fundamental research meant to be open and, beside it, in the same buildings and on the same networks, projects that are anything but. Export-controlled technical data under ITAR and EAR. Dual-use research of concern. Defence and national-security grants with security controls written into the award. Industry-sponsored work bound by confidentiality agreements that can void the funding if a byte leaks. Human-subject data, genomic data, patient cohorts. The compliance surface is not one regime. It is a dozen overlapping regimes stacked on the same infrastructure.

Generative AI is now central to that work. Researchers want to summarise literature, draft grant sections, analyse datasets, write and debug code, and interrogate their own findings in natural language. But the default way to get that capability is to send text and data to a public-cloud model provider, and for a large share of institutional research that single act is a breach. Export-controlled technical data leaving the country, even to a datacentre, can be a violation. Sponsor-restricted data sent to a third-party model can void the grant. The institution is left choosing between productivity and compliance, and research offices end up policing shadow AI use they cannot see.

What sovereign AI means for a campus

Mickai removes the choice. Mickai is a sovereign AI operating system, a SIOS: capable AI that an institution owns and runs entirely inside its own walls, on-premises and air-gapped. No prompt, no dataset, no intermediate result leaves the boundary. There is no external inference endpoint, no vendor telemetry, and no dependency on a public model API. The models run on hardware the university controls, in a facility the university already secures for controlled research.

That is the architectural answer to export control. If the technical data never leaves the physical and network boundary of the institution, there is no export event to license or to breach. A Technology Control Plan can list the AI system as an in-boundary resource rather than an uncontrolled leak. Sponsor confidentiality clauses are honoured by construction, because the confidential data is only ever processed by infrastructure the sponsor agreement already contemplates. The institution gets frontier-grade assistance for its researchers without moving a single controlled byte off-site.

A record the research office and the sponsor can trust

Compliance is not only about keeping data in. It is about being able to prove what happened. Mickai writes every action, every prompt, every retrieval and every model output to an Object Attestation Record, the OAR: a tamper-evident, post-quantum-signed audit log. Each entry is sealed with ML-DSA-65, a signing scheme chosen so the record stays verifiable against future quantum decryption, and bound to hardware-anchored identity so an action cannot be forged or repudiated.

For a research institution this changes the conversation with funders and regulators. When a defence sponsor, a research security office or an auditor asks who accessed a controlled dataset, what the AI was asked, and what it returned, the answer is a signed, immutable record rather than a reconstruction. Retrieval runs against an air-gapped RAG index built only from the institution's own corpus, so answers are grounded in sanctioned material and every citation is traceable. That is the evidentiary spine research security frameworks increasingly demand, produced automatically rather than assembled by hand at audit time.

Built for the people accountable

Responsibility for research security does not sit with one person, and Mickai is built for all of them. The CISO gets a system with no external attack surface for exfiltration and a cryptographic ledger of use. The research security officer and the export control officer get a demonstrable in-boundary control that maps to ITAR, EAR and Technology Control Plan obligations. The Data Protection Officer gets special-category and human-subject data that never leaves the estate, simplifying the GDPR DPIA. The General Counsel and the governing council get provable adherence to sponsor terms and a defensible position if a dispute ever arises. Each of them can point to the same signed record.

Under the hood, fifty specialist brains operate beneath a deterministic arbiter, so outputs are reproducible and governed rather than a black box. Where an automated action needs to be undone, compensating rollback reverses it cleanly. For the most sensitive operations, voice-biometric quorum can require named individuals to authorise together. This is the same architecture our Greek-named Studios run for regulated industry, from Nomos for compliance to Aletheia for audit and Panacea for clinical work, now brought to the campus.

Momentum, and a narrow window

Mickai is built and LIVE, and we are building to scale. The IP position sits behind it: 104 filed UK patent applications covering roughly 2,340 claims, held by Mickai LTD, establishing a priority and prior-art moat around sovereign, auditable, air-gapped AI. As a third-party momentum signal, in June 2026 I was verified at number four on Crunchbase, with the company placing in the top one to two percent globally. We are a UK company with manufacturing secured in Birmingham. Mickai is an ally to the wider AI ecosystem, not a challenger to it: we make frontier-grade capability usable where the public cloud legally cannot go.

The sovereign AI market is moving from around USD 40 billion in 2025 toward an estimated USD 148 billion by 2032, and research institutions are among the clearest cases for it. We are opening the current window to a small number of selected university and laboratory partners who want controlled-research AI running inside their own walls, not a pilot bolted onto someone else's endpoint. If that is your institution, I would like to talk.

Contact: micky@mickai.co.uk

Frequently asked questions

Can universities use AI on export-controlled or ITAR/EAR research data?

Not with public-cloud AI, because sending controlled technical data to an external provider can itself be an export event or a breach. Mickai runs entirely on-premises and air-gapped inside the institution's own boundary, so the controlled data never leaves and there is no export event to license. The AI system can be listed as an in-boundary resource on a Technology Control Plan.

How does Mickai satisfy sponsor confidentiality and research security requirements?

Confidential and sponsor-restricted data is only ever processed by infrastructure the institution controls, so the terms are honoured by construction. Every prompt, retrieval and output is written to the OAR, a tamper-evident, post-quantum-signed audit record bound to hardware-anchored identity, giving research security offices and sponsors a signed, immutable account of who accessed what and what the AI returned.

Does using AI still make research reproducible and auditable?

Yes. Fifty specialist brains operate beneath a deterministic arbiter, so outputs are governed and reproducible rather than a black box. Retrieval runs against an air-gapped RAG index built only from the institution's own corpus, so every answer is grounded in sanctioned material with traceable citations, and every action is captured in the OAR for later audit.

Is Mickai available, and how do institutions engage?

Mickai is built and LIVE, and we are building to scale. We are opening the current window to a small number of selected university and laboratory partners who want controlled-research AI running inside their own walls. Contact micky@mickai.co.uk to start a conversation.