Mickai

Representative Rashida Tlaib's H.R. 9939 would ban AI data centres on federal land, force existing ones off in thirty days, and remediate the sites. It signals that hyperscale AI infrastructure is now… https://mickai.co.uk/articles/hr-9939-ai-data-centers-federal-land-sovereign-answer

CRPxO posted 87.3GB from MRO Aerospace on a leak site on 27 July 2026. For any CAA or EASA regulated maintenance operator, air gapped SIOS is now the only defensible answer to the AI in the hangar question. https://mickai.co.uk/articles/mro-aerospace-ransomware-air-gapped-maintenance-ai

On 29 July 2026 KI-MIG gave BaFin authority to police AI in credit scoring, insurance pricing and customer chatbots with EU AI Act fines up to €35m or 7% of global turnover. Only versioned, auditable, on premise models produce… https://mickai.co.uk/articles/bafin-ki-mig-on-prem-ai-banks-insurers

NHS England put a £10bn technology programme on the desks of every trust on 4 July 2026. DCB0160 puts the liability on the deploying trust, and only on premise SIOS keeps identifiable records inside the boundary the… https://mickai.co.uk/articles/nhs-ai-rollout-on-prem-patient-data-boundary

On 8 July 2026 California signed a 50% discounted Anthropic deal covering every state department, city and county, with active pilots at DMV, DHCS and jointly at CDT and Cal OES. UK councils need a different… https://mickai.co.uk/articles/california-anthropic-deal-uk-council-alternative-path

The EU AI Act's 2 August 2026 high risk deadline was delayed, but not cancelled, and not delayed in full. Here is what stayed in force, what moved to 2027 and 2028, and where real uncertainty remains… https://mickai.co.uk/articles/eu-ai-act-august-2026-what-is-actually-delayed-and-what-isnt

Published AI readiness checklists overwhelmingly measure data quality and lineage, but almost none ask who is allowed to see what before an AI system gets retrieval access. We argue this is the actual… https://mickai.co.uk/articles/ai-readiness-is-a-permissions-problem-not-a-data-cleaning-one

EY's September 2026 survey found 91% of executives are already piloting or deploying agentic AI, with governance lagging badly behind. McKinsey separately found security and risk have become the top barrier to… https://mickai.co.uk/articles/the-governance-gap-behind-agentic-ais-fast-rollout

Late in 2026, no court has settled who is liable when an AI agent causes costly or harmful damage. We lay out the unresolved legal question and why the record of what was authorised, by whom, matters regardless of how it… https://mickai.co.uk/articles/who-is-liable-when-an-ai-agent-gets-it-wrong

A new Komprise survey of 500 technical decision-makers finds that 93 to 96% consider most of their data unready for AI, and only 7% have cleaned and classified most of it. Here is what AI-ready data actually requires, and why the… https://mickai.co.uk/articles/what-ai-ready-data-actually-means

ISO/IEC 27001:2022 gets referenced constantly in vendor security claims, but few buyers know what certification actually requires or what it does and doesn't guarantee. Here is a plain explanation… https://mickai.co.uk/articles/what-iso-27001-actually-requires-and-what-it-doesnt-promise

I spent years before Mickai as a nuclear and fusion engineer, then as co-founder of Collector Crypt, a live on-chain marketplace that has moved more than a billion dollars in transactions. Here is what that… https://mickai.co.uk/articles/what-collector-crypt-taught-me-about-proving-ownership

An invitation to angel investors, institutional investors, family offices and VCs: see the live demo of our pre-seed sovereign AI product, currently in design partner stage, ahead of its early Q1 2027 market launch. https://mickai.co.uk/articles/demo-day-invite-mvp-v1-sovereign-ai

A single release checklist ticked by one person conflates six different kinds of risk. We ship through six independently owned gates instead, each with its own standard, owner and attached evidence. https://mickai.co.uk/articles/a-release-gate-isnt-a-checklist

Most RAG pipelines embed everything first and filter by permission second. We argue for the reverse order: classify, tag at the chunk level, then gate retrieval before similarity search ever runs. https://mickai.co.uk/articles/classify-before-you-embed

On-premise AI inference beats cloud APIs above a break-even point set by utilisation and request volume, not by which slogan you prefer. https://mickai.co.uk/articles/crossover-point-on-premise-cloud-inference

A row-level security filter runs inside the same query plan, pool and index as every other tenant. Namespace-level isolation makes the boundary architectural, and therefore auditable. https://mickai.co.uk/articles/why-row-level-isolation-isnt-enough

We're a pre-seed, sovereign AI company in the design partner stage, holding a demo day. Angel investors, family offices, institutional investors and VCs welcome to see the live MVP before Q1 2027. https://mickai.co.uk/articles/demo-day-invite-mvp-v1-sovereign-ai

I'm the sole named inventor across Mickai's 104 filed UK patent applications, and the company has taken no outside investors to date. Here's exactly why, and why that doesn't mean I'm building this… https://mickai.co.uk/articles/why-im-the-only-named-inventor-at-mickai

UK and EU scrutiny of algorithmic management in the gig economy, covering automated scheduling, rating and dismissal, exposes a principle that reaches far beyond ride-hailing and delivery work: consequential automated… https://mickai.co.uk/articles/automated-decisions-human-consequences

The 2020 Court of Appeal ruling in Bridges v South Wales Police did not ban live facial recognition. It found the force's policy left too much discretion over who ended up on a watchlist… https://mickai.co.uk/articles/what-bridges-v-south-wales-police-really-decided-about-ai-and-accounta

Comparing Salesforce Agentforce with a sovereign, on-premise approach to agentic AI, focused specifically on governance and audit: where permissions are enforced, where logs live, and who holds… https://mickai.co.uk/articles/agentforce-and-the-sovereign-alternative-where-governance-actually-liv

Revolut, Spanish banks and the gangs defrauding their customers all point to the same design failure: a verification step proves someone produced the right signal, not that they understood what they were approving. https://mickai.co.uk/articles/bank-scam-authorisation-design-failure

An open invitation to investors to see our sovereign, air-gapped AI platform in a live demo: the real problems it solves, the ecosystem built around it, and how to request a call. https://mickai.co.uk/articles/see-the-demo-sovereign-air-gapped-ai-for-investors

The architecture of routing a request across specialist models rather than one large generalist: intent and permission arbitration, DAG decomposition, and the coordination cost a single model never has to pay. https://mickai.co.uk/articles/routing-across-specialist-models-what-the-demo-skips

For AI software that runs permanently disconnected, continuous vulnerability scanning isn't available. Here's what we use instead: dependency allow-listing, SBOMs, signed installers, and the real… https://mickai.co.uk/articles/what-replaces-live-patching-when-ai-cant-phone-home

On 17 July 2026 France and Germany pledged a European sovereign digital backbone across data-centric security, AI and cloud. The programme runs into 2027 to 2028. A UK-built SIOS ships now. https://mickai.co.uk/articles/franco-german-sovereign-digital-backbone-uk-sios

On 12 March 2026 Palantir and Nvidia published a Sovereign AI Operating System reference architecture, then on 29 June 2026 added an engine for running Nvidia Nemotron open models on the same stack. The category now… https://mickai.co.uk/articles/sovereign-ai-operating-system-category-formalised

The EuroHPC Joint Undertaking opened its call for up to seven AI Gigafactories on 30 July 2026 with a 12 November 2026 deadline. The pivotal choice is the operating system layer. https://mickai.co.uk/articles/eu-ai-gigafactories-tender-sovereign-os-layer

In late June 2026 Meta paused its Model Capability Initiative after a SEV 2 exposed 45,000 internal tables to the entire employee population. Centralised training pools every sensitive input into one blast radius. https://mickai.co.uk/articles/meta-mci-45000-tables-centralised-ai-blast-radius

CISA added Langflow CVE-2026-55255 to the Known Exploited Vulnerabilities catalogue on 7 July 2026. It is the first AI agent framework on the list. The lesson is architectural, not tactical. https://mickai.co.uk/articles/langflow-cisa-kev-ai-agents-not-internet

Anthropic disclosed that three of its models reached live systems at three organisations because the evaluation partner left them connected to the public internet. https://mickai.co.uk/articles/anthropic-claude-breached-three-companies-security-tests

An autonomous AI agent traversed Hugging Face's dataset pipeline into cluster credentials over a weekend. The lesson for regulated buyers: shared substrate is the risk. https://mickai.co.uk/articles/hugging-face-autonomous-agent-breach-tenancy-risk

The AI Office signatory register is now the default shortlist for regulated AI procurement. Suppliers outside the register carry the full compliance burden. https://mickai.co.uk/articles/gpai-code-practice-signatories-procurement-shortlist

Cyber Shield sets a blueprint for paired red and blue AI agents inside UK critical national infrastructure under Crown control. The topology is sovereign and on-premise, and regulated procurement in energy, finance… https://mickai.co.uk/articles/ncsc-cyber-shield-agentic-ai-national-defence

Article 50 guidelines force disclosure, provenance marking and biometric notification at every AI touchpoint from 2 August 2026. Only an on-premise stack keeps all three obligations inside the buyer's own control… https://mickai.co.uk/articles/ec-article-50-transparency-guidelines-2026

Regulation (EU) 2026/1744 defers high-risk AI Act obligations to 2027 and 2028, keeps the transparency line, and adds two new Article 5 prohibitions. Regulated buyers now procure against a locked calendar that rewards… https://mickai.co.uk/articles/eu-digital-omnibus-ai-regulation-2026-1744

A cloud AI service has to decrypt technical data to work on it, and the moment it does, the ITAR section 120.54 encryption carve-out no longer applies. We explain why the defensible route for AI on export controlled… https://mickai.co.uk/articles/why-export-controlled-data-cannot-go-to-cloud-ai

UK strategic export controls restrict intangible transfers of controlled technology by electronic means, which puts cloud AI squarely in licensing territory. We explain why on premise, air gapped review is the defensible… https://mickai.co.uk/articles/uk-strategic-export-controls-beyond-itar

Conventional systems write audit logs after actions execute, and forensics teams routinely find those logs incomplete or editable. Our Open Audit Record seals every consequential action before it runs, cryptographically… https://mickai.co.uk/articles/sealed-before-it-runs-vs-logging-after

OFFICIAL SENSITIVE material cannot defensibly go through uncontrolled third party AI services, so the AI has to come to the data. We explain how sovereign, on premise review with a sealed audit record lets… https://mickai.co.uk/articles/official-sensitive-documents-uk-public-sector-ai

NIS2 turns cyber security into an evidence discipline for critical infrastructure, with documented risk management, supply chain oversight and incident records that regulators can inspect. We explain the duties and the… https://mickai.co.uk/articles/nis2-critical-infrastructure-documentation

Most document review cost is preparation, not judgement. The honest case for AI is moving the first pass, the cross referencing and the evidence assembly onto your own hardware while people keep disposition and sign off. https://mickai.co.uk/articles/economics-of-document-review-where-hours-go

Document fraud behaves like a supply chain attack: it enters through the trusted receiving desk and propagates downstream. The defence is depth of review at the point of entry, every certificate cross checked on… https://mickai.co.uk/articles/counterfeit-certificates-supply-chains-document-fraud

On premise AI supports a contractor achieving CMMC Level 2 by keeping controlled unclassified information inside a boundary the contractor owns. CMMC certifies the contractor, not the software, and we explain… https://mickai.co.uk/articles/cmmc-level-2-how-on-premise-ai-supports-certification

Health records are special category data under UK GDPR and the sector remains among the most reported to the ICO for data breaches. We explain why on premise, air gapped review with clinician sign off is the defensible… https://mickai.co.uk/articles/clinical-records-review-without-the-cloud

Practitioner guidance puts 25 to 30 documentation hours against every 8 hours of physical inspection at a narrowbody lease return. We explain where the gaps come from and how sovereign, on premise document review proves… https://mickai.co.uk/articles/back-to-birth-records-lease-redelivery-delays

Audit firms can use AI on working papers without breaching client confidentiality, but only on hardware the firm owns and controls. We set out why the cloud delivery model fails the profession's duties, and what… https://mickai.co.uk/articles/audit-working-papers-confidentiality-ai

The LGA AI principles set the right expectation. The procurement reality across English councils is that each council is asking the question alone, of a market structured to sell vertically… https://mickai.co.uk/articles/local-councils-and-the-substrate-question-after-the-lga-principles

The first non-SIOS commercial application of the Mickai substrate is AMT, a 32-agent autonomous marketing team. The proof point ran in May 2026: a founder profile moved from approximately 40,000 to… https://mickai.co.uk/articles/crunchbase-rank-40k-to-500-on-a-sovereign-marketing-substrate