The UAE Merged Three AI Regulators Into One. Most Companies Won't Merge Two.

Five days after our last briefing, the honest answer to "have things slowed down" is no — and the honest answer to "are companies still fighting silo fires" is also no different: yes, badly, and the data this week says it's getting worse, not better, even as the headlines keep accelerating. The most interesting story of the week isn't a new model or a new law. It's that the UAE government just made a structural choice most companies won't make: it merged three separate regulatory bodies overseeing AI and data into one authority, rather than adding a fourth committee to coordinate between them. That's not the same as saying the UAE's underlying data systems are now unified — nothing in this week's reporting claims that — but it is a real, verifiable example of treating fragmentation as an organisational-design problem rather than a policy footnote, which is precisely what 82% of C-suite executives say their own companies still haven't done. Here's what happened, verified, region by region — UAE first — and what the silo data means for how you plan the rest of this quarter.

Key Takeaways

UAE — a regulatory merger at the top, real teeth at the bottom

Two UAE stories this week point in the same direction from opposite ends of the system. At the top, the US Commerce Department's decision to ease export controls — reported this week as the UAE is reclassified as a more trusted partner for advanced computing technology — is being read by regional press as a direct reward for the UAE's support during US operations against Iran, following years of Emirati lobbying for broader chip access. It's a geopolitical unlock, not a technical one: the ceiling on what the UAE can now legally import just moved.

At the bottom, enforcement is arriving. Sheikh Mohammed bin Rashid Al Maktoum's Federal Authority for Artificial Intelligence and Data — announced last month, folding the Office of AI, Digital Economy and Remote Work Applications, the TDRA's Digital Government Sector, and the UAE Data Office into a single body under Omar Sultan Al Olama — is now the first fully operational supervisor of the UAE's Personal Data Protection Law since it came into force in 2022. Speaking to AGBI this week, Denodo's Gabriele Obino described the problem this kind of authority is meant to address: "Most large organisations run dozens or even hundreds of systems, each holding a slightly different version of the same business information. That creates genuine uncertainty over which version should be feeding an AI decision at any given moment." To be precise about what the UAE has and hasn't done: it has not claimed to have solved that underlying data-fragmentation problem, inside or outside government. What it has done is remove the simpler, prior failure mode — three separate regulators with three separate mandates and no single owner. That's a lower bar than "fixed," and worth naming as exactly that.

"Most companies won't even merge two regulators, let alone three. The UAE didn't publish another AI policy this week — it removed a coordination problem by deleting it, rather than managing around it. Whether that translates into better data governance in practice is the thing to watch next, not something to assume."

— Lisa Warren, Founder & CEO, Neural Horizons AI

Meanwhile, AI Appreciation Day brought a chorus of Gulf-based tech leaders to the same conclusion from a different angle. Dr Moataz BinAli of Magna AI said the region has moved from importing AI to "building, owning, and shaping" it through local infrastructure and IP. ServiceNow's Antonio Rizzi pointed to the rise of "agent loops" — autonomous systems that pursue goals, coordinate with other agents, and refine their own outcomes — as 2026's defining shift, moving people from being "in the loop" to being "on the loop." AppsFlyer's Inna Weiner and Anomali's Alexandre Depret-Bixio both landed on the same warning from different directions: an AI system is only as trustworthy as the governance and data quality behind it, and poorly governed autonomy fails at machine speed, not human speed.

The wider GCC — adoption is real, the ecosystem is still being built

Saudi Arabia's Communications, Space and Technology Commission put a hard number on where the Kingdom actually stands: 45.2% of Saudi internet users have adopted AI tools, according to its Internet 2025 report released this week. That's a meaningfully different kind of data point than the infrastructure and contract announcements that dominated last week's coverage — it's the demand side, not the supply side, and it suggests the Kingdom's aggressive infrastructure build-out has a real user base to serve.

Riyadh kept up the diplomatic pace too: Minister Abdullah Alswaha met Korean firms this week to pursue partnerships across AI, semiconductors, digital infrastructure and cloud computing, and separately held talks with Chinese government and technology leaders to expand cooperation on AI, the digital economy and talent exchange. The direction of travel is consistent with what we covered last week — sovereignty through partnership, not isolation — but this week's news is about who Saudi Arabia is now doing that with, not just how much it's spending.

United States — the China question gets sharper, the state patchwork keeps growing

The sharpest story out of Washington this week has nothing to do with domestic AI law. Treasury Secretary Scott Bessent said the administration could sanction Chinese AI labs if it confirms they built their models by stealing from American ones — specifically citing "distillation," where a smaller model is trained on a larger model's outputs. "We are finding watermarks of our U.S. large language models on many of the Chinese models, and that's unacceptable," Bessent said. The claim isn't abstract: Anthropic has separately reported that operators linked to Alibaba ran a mass-distillation campaign against Claude, generating more than 28.8 million interactions through roughly 25,000 fraudulent accounts between April and June. The timing is notable — this surfaces just ahead of the first official US-China AI dialogue under the current administration, expected in September, before President Xi's planned US visit later that month.

Domestically, the FTC's proposed policy statement deserves a more precise description than "accuracy standards." Issued July 1 at President Trump's direction, it addresses AI companies allegedly "manipulating the behavior of their AI systems" for "undisclosed ideological objectives," and explicitly argues that state laws compelling AI output changes — Colorado's AI Act is named directly — may be federally pre-empted. Chairman Andrew Ferguson framed the goal as advancing "President Donald Trump's goal of expanding America's global dominance in artificial intelligence." This is a federal-versus-state fight over who gets to define AI "neutrality," not a technical standards process, and public comment stays open through July 31 for anyone who wants to weigh in before it hardens.

Separately — and on a genuinely different track — the state-by-state patchwork on AI in healthcare keeps expanding. Alabama's SB 63, enacted in April and taking effect October 1, requires that any decision to deny, delay or modify a prior-authorization request based on AI input be made by a licensed clinician, not the model. Georgia's SB 544, effective January 1, 2027, sets a similar human-review requirement for AI-influenced coverage denials. Neither is new this week, but both are worth flagging now: the compliance clock on both is closer than the headlines suggest.

Europe — from deadline to instructions

Last week we flagged August 2 as the date the EU AI Act's transparency obligations go live, with no grace period for most requirements. This week, the European Commission moved from setting the deadline to explaining how to survive it: on July 20, it published detailed guidelines to help providers and deployers actually meet Article 50's labelling and disclosure requirements. If your compliance team has been waiting for clarity rather than just a countdown, it's now available — with less than two weeks left to act on it.

China — building around the chip gap, and the market notices

Z.ai, the Beijing-based lab formerly known as Zhipu, has partially activated a roughly 1-gigawatt data centre — reported by Bloomberg on July 20 — running entirely on Chinese-made chips, with several clusters of more than 10,000 chips each trained on Huawei's Ascend accelerators and MindSpore software. Worth being precise about why: the US Commerce Department blacklisted Zhipu in early 2025, cutting it off from legally purchasing Nvidia silicon. This is an infrastructure response to having no other option, not a preference — and Chinese accelerators, including Ascend, still trail Nvidia's current generation on per-chip performance and energy efficiency. The market is pricing in the resilience of the attempt, if not yet full parity: AI-linked shares are the one segment of China's $13.5 trillion stock market that has performed strongly this year, with telecom and tech gauges on the CSI 300 up more than 60% in the first half of 2026, even as the broader index lags.

Governance: the silo data, in full

Here's the direct, sourced answer to this week's question. Adoption has not slowed down — if anything, the pace of deployment keeps outrunning the ability of organisations to govern it. But "not slowed down" is not the same as "not siloed." Two separate 2026 studies converge on the same conclusion from different angles. WRITER's survey of 2,400 executives and employees found 79% of organisations report AI adoption challenges, a double-digit increase over 2025 — and, in a distinct finding from the same study, 79% of executives separately say their own AI applications are being built in silos right now, 53% say growing tension between IT and other business lines means IT isn't delivering real value, and 55% describe AI use inside their company as a "chaotic free-for-all." More than half — 54% — of C-suite leaders admit adopting AI is "tearing their company apart," despite 59% of companies spending over $1 million a year on it. Only 29% see significant ROI from generative AI, and just 23% from AI agents specifically.

IBM's Institute for Business Value, surveying 2,000 senior leaders across 16 countries for a report covered by The National CIO Review, puts a number directly on the structural problem: 82% of C-suite executives say functional silos are actively blocking AI value.

The pilot data tells the same story from a different angle. Astrafy's synthesis of the research puts it plainly: 67% of AI projects fail to become a real business asset. McKinsey finds nearly two-thirds of organisations still stuck in pilot mode, unable to scale across the enterprise. BCG's numbers are starker — 60% of companies report "hardly any material value" from their AI investment, and 74% have yet to show any tangible value at all. On the technical side, Gartner attributes 85% of AI project failures to poor data quality, and separately forecasts that 30% of generative AI projects get abandoned entirely after proof-of-concept.

There is one genuinely encouraging number in the data: 55% of organisations, per IBM's research, are now developing or deploying an agentic AI operating model specifically designed to break down departmental boundaries — meaning more than half the market has at least correctly diagnosed the problem, even if most haven't solved it yet.

What to do, concretely

• Don't wait for a reorganisation to fix your silo problem. The UAE's move is a useful example of choosing structural ownership over another cross-functional committee — not proof that the approach alone solves fragmentation. Find the one accountable owner for AI-fed data across departments, even if that means real org-chart change, and treat it as day one of the work, not the end of it.
• Before greenlighting another pilot, ask who owns it in production, not just in the sandbox — Gartner's data says unclear ownership, not model quality, is what kills most pilots at the production boundary.
• If your compliance team is targeting August 2 for the EU AI Act, use the Commission's July 20 guidance now — it answers the "how," not just the "by when."
• If you operate in the UAE, treat the Federal Authority for AI and Data as a live regulator starting today, not a policy body — the Personal Data Protection Law now has real enforcement behind it for the first time since 2022.

What leaders should take away this week

The first takeaway is that capability and geopolitics are now the same story — the US eased chip export controls for the UAE as a direct consequence of a security relationship, not a technology assessment, and China's Z.ai data centre exists because of a restriction, not despite it.

The second is that enforcement is arriving faster than most companies are ready for — the UAE's new authority went from announcement to active supervision in five weeks, and Europe's Article 50 deadline is now a matter of days, not a future date.

The third, and the one worth sitting with, is that the silo problem is not an AI problem — it's an organisational design problem that AI is now making visible and expensive at the same time. 82% of C-suite executives can already see it. The UAE government just showed what choosing to act on that diagnosis looks like at the regulatory level — a real, if partial, data point, not a finished case study.