Unified ยท A Mickai journal
Guidance on sovereign AI, audit and regulation for public-sector bodies, regulated industries, and the procurement and risk teams accountable for the AI they choose.
Latest article
· 5 min read
Written for
Published by
Mickai LTD, a British company building a Sovereign Intelligence Operating System. Written by Micky Irons.
About the journal and teamThe journal
Short, direct answers to the questions public bodies and regulated organisations ask about AI they can control. Browse by topic.
Explainer
Model Context Protocol (MCP) is quietly becoming the standard way AI agents plug into company data and tools. That convenience is exactly why it needs auditing: every MCP server is a new door into the organisation, and most of them have never been checked.
Analysis
A single misconfigured setting in a popular chatbot app exposed hundreds of millions of private conversations, including disclosures of self-harm. The incident is a useful test of what sovereign or on-premises AI actually fixes, and what it does not.
Guidance
Most vendors say their AI is auditable. Fewer can answer the actual question: can you check the record is genuine without them, without the internet, and without trusting their word for it?
Perspective
Trust Agent is Mickai's free AI education site: more than sixty no-login PDF workbooks from your first prompt to frontier models. Why it is staying free, and what that has to do with trust.
Explainer
DeepSeek moved to V4, Llama moved to 4.5, and reasoning stopped being a separate model family. Here is what actually shifted, and why it matters if you are choosing what to run on your own hardware.
Compliance
NIS2 does not have an AI clause. It has a supply-chain security clause, and an AI supplier sits inside it like any other. Here is what that actually requires.
Explainer
Trust Agent's downloadable AI agents, a code reviewer, a project manager, a research assistant and more, explained plainly: what makes them different from a chatbot, why they cost nothing during beta, and why that is not a promise forever.
Perspective
Free AI learning from Mickarle Wagstaff-Irons - Micky Irons: browser labs, practical study and a first software project.
Compliance
The Digital Operational Resilience Act treats an AI vendor like any other ICT third party. Here is what that actually requires, and why it is simpler when the AI runs on infrastructure you control.
Compliance
CSMv4, DEFCON 658 and Def Stan 05-138 do not have a separate track for AI. An AI supplier faces the same risk-based assessment as any other, with one added difficulty: showing what the system actually did.
Analysis
Explore tokenisation, temperature, top-k and top-p with worked examples, and see what browser experiments cannot prove about a real AI system.
Analysis
Map a fictional document assistant's trust boundaries, abuse cases and evidence gaps before choosing controls or admitting real files and tool actions.
Analysis
Calculate weights, KV cache and headroom for a fictional local AI workload, then separate a memory estimate from evidence that a private pilot will work.
Analysis
Use a small numerical example and a paired evaluation plan to compare quantised models on task failures, runtime behaviour and memory, not file size alone.
Analysis
An AI learning roadmap for non-coders: choose a useful task, practise with fictional material and check your work before advancing.
Explainer
Sovereign AI means running artificial intelligence entirely under your own control, on infrastructure you own, with nothing forced through an outside provider. Here is what that means in practice, and why regulated organisations are moving toward it.
Explainer
They sound similar and vendors use them loosely. Here is the plain distinction, and why it decides whether your data is genuinely under your control.
Explainer
Yes. Modern AI can run with no internet connection at all, on hardware you own. Here is how air-gapped AI works, what it can and cannot do, and who needs it.
Guidance
For most regulated work the honest answer is no, or not without real care. Here is what actually happens to your data, and the safer alternatives.
Compliance
High-risk AI systems have to keep records of what they did. Here is what the logging and record-keeping duties mean in practice, and why they are easier to meet with AI you control.
Explainer
An AI audit trail records what an AI system did, so it can be checked later. Here is what a good one looks like, and the difference between tamper-evident and 'tamper-proof'.
Guidance
Twelve questions that separate genuine sovereignty from private-cloud marketing. If a vendor cannot answer these plainly, that is your answer.
Landscape
Sovereign AI is a real field with serious participants. Here is how to think about the landscape, and where a British company is pushing past the baseline.
Explainer
Yes. Useful AI now runs on workstation and server-class hardware a business can own outright. Here is what it takes, and where the limits are.
Explainer
The phrase is new and often loose. Here is a working definition, and how an AI operating system differs from a chatbot bolted onto the tools you already rent.
Profile
Mickarle Wagstaff-Irons - Micky Irons, full name Mickarle Sean Junior Wagstaff-Irons, is the founder of Mickai, a British company building a Sovereign Intelligence Operating System. Here is the short version.
Try a broader word, or clear the filters to see the whole journal.
Start here
A short reading path for anyone briefing a board, writing a requirement or assessing a supplier. Each piece answers its question directly, then explains.
The definition
The distinctions
The accountability record
The buying questions
Who writes Unified
Unified is published by Mickai LTD, a British company building a Sovereign Intelligence Operating System, and written by Micky Irons. It covers the field broadly and treats Mickai as one example within it. Where Mickai is discussed, the relationship is stated on the page.
