Field notes from inside a sovereign AI.
Mickai® is a Sovereign Intelligence Operating System (SIOS) that runs on your own hardware. These are long-form essays on the architecture of the Mickai SIOS, the patterns that keep surfacing in commercial AI in 2026, and the engineering choices that make sovereign intelligence possible. Written by Micky Irons, named inventor of the 104 filed UK patent applications, recorded on the UK IPO public register at numbers GB2607309.8 to GB2611702.8, GB2611885.1 onwards, GB2612762.1 to GB2612793.6, GB2613386.8 to GB2613404.9, and GB2615041.7 to GB2615043.3.
The cooperative the field notes describe, running on the sovereign silicon substrate.
Showing 145 to 168 of 1124 articles.
DORA and the document trail your firm must be able to prove
DORA turns operational resilience into a documentary obligation, and the firms that can produce a provable trail of registers, contracts and incident records on demand are the ones that will face inspection calmly.
The economics of document review: where the hours actually go
The hours in document review go on first pass reading, cross referencing and evidence assembly, mechanical work a sovereign system can carry while people keep judgement and sign off.
Due diligence data rooms and AI: privilege is the constraint
Law firms can use modern AI on a due diligence data room without endangering privilege only when the review runs on their own hardware, air gapped, with every action sealed to a verifiable record.
Half of MRO engineering time goes on paperwork. It does not have to.
Industry commentary puts half of MRO engineering time on paperwork, and the fix is a sovereign first pass on your own hardware with engineers keeping sign off.
NIS2 and the documentation burden on critical infrastructure
NIS2 obliges essential and important entities to document cyber risk management, supply chain security and incident handling, and to prove it on demand, with management personally accountable.
OFFICIAL SENSITIVE and AI: the question UK departments must answer
The defensible way for a UK department to apply AI to OFFICIAL SENSITIVE material is to run it inside its own estate, air gapped, with every action sealed and a person in charge.
Part 145 record keeping: from asserted compliance to provable
Repair stations become provably compliant when every maintenance record is sealed, signed and verifiable offline at the moment the work happens.
Pharma batch records: Part 11 wants exactly this audit trail
Part 11 requires secure, computer generated, time stamped audit trails independent of the operator, and a sovereign on premise review system with a sealed, tamper evident record is the cleanest way to meet it while using AI on batch records.
Sealed before it runs beats logged after it happened
An audit record sealed before an action runs cannot be shaped by the outcome, which makes it evidence rather than recollection.
Supplier certificates: manufacturing's quiet counterfeit problem
Counterfeit parts enter factories through paperwork, and deep cross referencing of every supplier certificate at receiving, on your own hardware, is the defence that works.
UK Export Controls and Cloud AI: The Intangible Transfer Wall
Uploading controlled technical data to a cloud AI raises a licensable intangible transfer question under UK export control law, and sovereign on premise review removes it.
Why export controlled technical data cannot go to cloud AI
Cloud AI must decrypt technical data to process it, which breaks the ITAR encryption carve-out and turns provider staff into deemed export exposure, so on premise, air gapped review is the defensible route.
Banks outpace regulators on AI, and the assurance gap must close
The fastest way to close the gap between AI adoption and AI assurance is to make every model action produce its own signed, verifiable evidence as it happens.
Colorado's revised AI law is a documentation test worth passing
Senate Bill 26-189 turns AI compliance into a question of evidence, and organisations that record decisions as they happen will be ready before 1 January 2027.
ITAR and CMMC Bite in 2026: How UK Defence Suppliers Keep Controlled Data Sovereign and Stay Contract Eligible
Keep controlled technical data on hardware you own, produce the CMMC and DEFCON 658 evidence, and drop the cloud SIEM and GRC subscriptions.
AI chatbot misuse tops ECRI's 2026 hazards. We argue for governed AI
ECRI named AI chatbot misuse the top health technology hazard for 2026 because unvalidated, unregulated chatbots already shape medical decisions, and the answer is governed clinical AI with consensus checks, clinician sign off and a verifiable audit record.
EU GPAI enforcement begins: what an audit trail must prove
From 2 August 2026 the European Commission can compel general-purpose AI providers to prove their claims, and a signed audit record is what proof looks like.
The EU AI Act deferral buys time, not a pass on auditable AI
Deferred deadlines change when the high-risk obligations bite, not what a regulator will eventually ask your AI to prove.
The EU buys sovereign cloud, and the AI layer needs the same logic
The European Commission's 180 million euro sovereign cloud award proves sovereignty can be procured, and the AI layer above it needs the same independence.
When AI breaks post-quantum crypto, agility is the real defence
Crypto-agility, not faith in any single algorithm, is what keeps post-quantum security standing when a scheme falls.
What the ICO's £963,900 fine means for provable accountability
The ICO's £963,900 penalty for South Staffordshire Plc shows regulators now expect organisations to prove their controls, and a sovereign, air-gapped operating system with a signed audit record is how that proof gets built.
Shared IT felled London councils. AI must not repeat the mistake
Councils reduce the blast radius of shared IT failures by running AI on hardware they own, air-gapped and audited, rather than adding another shared dependency.
AI medical devices go live in the NHS, the controls that make it safe
Live NHS deployment of AI medical devices is safe only when every output is provable, humanly overseen and held on infrastructure the trust controls.
The five hour Microsoft 365 outage and the case for sovereign AI
AI inherits the failure modes of wherever it runs, and on 23 July 2026 that meant around five hours of silence.