Thinking that holds up in the boardroom.

Samture adds AI governance and runtime control to its foundation for trusted AI in the UAE: one place to see which AI systems an organisation runs, set the rules they must follow, check them while they work, and show the evidence. Delivered under the Samture name, inside the UAE. Read the announcement.

Big Ambition Needs a Floor: Abu Dhabi will be the world's first AI-native government next year, Dubai wants half its services delivered by AI agents, and people's lives run straight across the lines between them. The UAE National AI Strategy 2031 calls the layer underneath the Foundation. What it does, shown for the first time in public on connected public services, an AI register that checks itself and a secure test bed, and why the same work keeps a bank's reasoning in the building when an experienced analyst leaves. Read the full edition.

The Sandbox Had a Door: during a security evaluation, Google's Gemini made its way into three real organisations, believing they were part of the test, and stopped by itself. The model did what it was asked. What gave way was a boundary that existed only as a word, and it took two months for anybody to notice. With Dubai moving half of its government services to AI agents, the one question to answer before go-live: what can it reach? Read the full edition.

The Answer That Looked Complete: the same question, five times, four different answers. Asked of the same model over the same documents, and every answer looked finished. What it takes to make an answer repeatable, sourced and defensible, measured on Samture's own business, including the control that reported a clean sheet across 188 agreements while having nothing at all to compare. Read the full edition.

AI's Last Mile: the money doubled, the foundation did not. Spending on AI in the Emirates rose 105% in a year while the average maturity score came out at 48 out of 100, and only 16% of organisations can test or audit what they are already running. That gap is not wasted money, it is the part of the work that never appears in a business case. Read the full edition.

The Mark Only Goes One Way: we can prove what the machine did, not what we did. Since 2 August, text generated by newer AI models carries an invisible watermark and the EU AI Act requires it. Machine involvement became provable; human involvement did not — and every retention policy you were right to write widens the gap. Read the full edition.

The Chair Nobody Is Sitting In: Dubai regulated a person, not a system. The DIFC has required a named Autonomous Systems Officer for high-risk AI since late 2023. Almost nobody has filled the seat — and most architectures could not give that person anything to sign for. Read the full edition.

The Speed Trap: the next phase of AI will not be won on speed. Boards and regulators no longer ask whether your AI works — they ask whether you can explain, govern and defend it. The real race is for legibility. Read the edition of The Signal & The Noise.

Signal and noise: what agentic AI actually changes in the enterprise. Autonomy is not the breakthrough. Accountability is — and most architectures cannot carry it yet.

Semantic intelligence: the domain almost nobody measures. Why organisations that skip the knowledge layer keep rebuilding the same AI project under new names.

The UAE just set a two-year deadline for agentic AI. Sheikh Mohammed’s directive to run half of government on agentic AI is not a government story — it is the business deadline of every organisation under the same compliance pressure.