AI Governance14 min readJuly 11, 2026

Geneva Says AI Can't Be Controlled. The Gulf Bets on Chips

The UN's first all-nations AI summit admitted no one can guarantee control over advanced systems. Riyadh and Abu Dhabi answered differently: more chips, new institutions, and a bet that governance can be built while racing ahead.

A Summit That Admitted the Obvious

For two days in Geneva, representatives of all 193 UN member states sat in the same rooms as technology executives, scientists and civil society groups and did something governments rarely do in public: they admitted they do not know how to control what they are building. The Global Dialogue on AI Governance, which opened on July 6 and closed the following day, was billed as the first intergovernmental gathering to bring every UN member into a single conversation about artificial intelligence. It did not produce a treaty or a binding rulebook. What it produced was more unsettling than that. A panel of forty scientists told the assembled diplomats, in writing, that no one currently has a technical guarantee that advanced AI systems will keep following the instructions they are given.

That sentence should not be controversial. Anyone who has watched a coding agent wander off script or a chatbot invent a citation already suspects it. But hearing it stated flatly by an independent scientific body, then repeated by the UN Secretary-General himself, changes the political weight of the claim. It turns an engineering caveat into a governance problem that 193 governments now have to answer, whether or not they have the computing power to do anything about it. Two of those governments, Saudi Arabia and the United Arab Emirates, spent the same week trying to answer it in the opposite direction: not by slowing down, but by building faster and writing their own rules as they go.

What separates this gathering from earlier summits that talked about AI without changing much is who was actually in the room. This was not a closed session among technology ministers from a handful of industrial powers. It brought delegates from countries that have never had a hand in writing a single global technical standard. Their presence does not hand them the same leverage as Washington or Beijing, but it means the absence of that leverage is now on the official UN record rather than a grievance raised quietly in a hallway.

What Forty Scientists Actually Wrote

The Independent International Scientific Panel on AI, chaired jointly by Yoshua Bengio and Maria Ressa, published its preliminary report on July 1, five days ahead of the summit. Bengio's warning was precise rather than alarmist: "science currently cannot guarantee that as capabilities continue to increase, AI will not cause catastrophic harm." Guterres opened the summit by underlining the point himself: "the science is here, we can no longer say we did not know." Ressa, whose work has focused on information integrity since her time running Rappler in the Philippines, framed the risk in narrower terms that many delegates found harder to dismiss. She called the current state of AI-driven disinformation an information emergency and warned that democratic institutions are already absorbing the damage.

The panel's language mattered because of who wrote it. This was not a lobbying document from a company with a product to sell or a competitor to slow down. It was assembled from researchers across every world region, working from published evidence rather than speculation about future models. Their central finding was less about doom and more about pace. Task complexity that AI agents can handle is doubling every few months, and systems capable of planning, using digital tools and writing software with limited supervision are already active in finance and commerce. Regulators, the panel argued, are working from an evidence dilemma: they need proof of harm before they can justify rules, but by the time that proof accumulates, the technology generating it has already moved twice.

Ninety Percent of the World's AI Sits in Two Countries

Strip away the diplomatic language and the summit's real subject was a map of computing power. According to the scientific panel's assessment, the United States controls roughly 75 percent of the supercomputing capacity behind the world's most capable AI systems, and China holds close to 15 percent. Together that leaves the other 191 UN member states sharing what is left, and most of them lack the independent capacity to even audit a frontier model, let alone train one. That single statistic explains why the Geneva dialogue drew so many finance ministers and digital-economy officials rather than only AI policy specialists. Governance without computing power is an opinion. Governance backed by infrastructure is a seat at the table.

Over 40 separate AI governance frameworks already exist worldwide, spanning national strategies, corporate voluntary codes and regional pacts, and the panel's report described them as fragmented and largely untested against real incidents. Layering a 42nd framework onto that pile changes little if the country writing it has no chips, no data centers and no leverage over the companies building the models it is trying to regulate. That is the gap Gulf states have spent the past two years trying to close, not through regulation first, but through infrastructure first.

Global AI Supercomputing Capacity by Country

Source: Independent International Scientific Panel on AI, preliminary report, July 1, 2026

The Chips That Bought the Gulf a Different Argument

Eight months before Geneva, on November 20, 2025, the US Department of Commerce authorized the export of up to 70,000 advanced Nvidia GB300 Blackwell chips split evenly between G42 in Abu Dhabi and Humain in Riyadh, 35,000 units each. The approval came wrapped in oversight: both companies operate under compliance regimes monitored by the Bureau of Industry and Security, with G42's version branded the Regulated Technology Environment. G42 chief executive Peng Xiao called it a move "from planning into execution," and Humain chief executive Tareq Amin thanked US officials for what he described as full export control approval for Nvidia's latest GPUs.

That deal is why the Gulf's presence at a summit about controlling AI does not read as contradictory to the officials involved, even if it looks that way from outside. Riyadh and Abu Dhabi are not asking to slow the frontier down. They are asking to be counted among the handful of jurisdictions that can build at the frontier, audit what runs on their own soil and negotiate compliance terms directly with Washington rather than wait for rules set elsewhere. Access to 35,000 Blackwell chips does not close the gap with 75 percent US and 15 percent Chinese compute, but it moves both countries out of the bucket of 191 states the scientific panel says cannot independently verify anything.

Advanced AI Chips Approved for Export, November 2025

Source: US Department of Commerce approval, reported by Middle East AI News, November 20, 2025

Riyadh Turns Up With an Institution, Not Just a Statement

Saudi Arabia sent Abdullah Al-Ghamdi, president of the Saudi Data and Artificial Intelligence Authority, to Geneva on July 8. His public remarks stayed close to the diplomatic script, affirming the kingdom's commitment to advancing international AI governance and to what he called the responsible and safe use of the technology in support of sustainable development. The more interesting part of his trip was who he met. Al-Ghamdi held separate sessions with the International Labour Organization's director general Gilbert Houngbo, World Bank vice president Sangbu Kim, Germany's federal minister for digital transformation Karsten Wildberger, and UNESCO director general Khaled El-Enany.

That itinerary lines up with an announcement Saudi officials had already been building toward: a new International Center for Artificial Intelligence Research and Ethics in Riyadh, operating under UNESCO's umbrella. The kingdom is also opening its doors in September, hosting the fourth Global Forum on the Ethics of Artificial Intelligence from September 14 to 17 as part of what it has designated its Year of Artificial Intelligence. None of this required new legislation. SDAIA had already published its own AI ethics principles back in 2025, well ahead of Geneva, so the Riyadh trip functioned less as a debut and more as an invitation for the rest of the world to plug into a framework Saudi Arabia had already written for itself.

Abu Dhabi's Bet: Govern the Outcome, Not the Model

Omar Sultan Al Olama arrived in Geneva a day after the Saudi delegation, on July 9, representing UAE president Sheikh Mohamed bin Zayed Al Nahyan. His framing of the UAE's position was more pointed than the general calls for cooperation heard elsewhere in the hall. Governance, he argued, should shift from regulating individual technology to governing its impact, and from scrutinizing single AI models to overseeing the ecosystems they operate inside. It is a subtle distinction, but it matters for a country that has already committed to running half of its government operations through agentic AI within two years, a target few other administrations have dared to put a number on.

The UAE backed that claim with two concrete builds rather than a promise. It has already stood up what it describes as the world's first AI-powered Regulatory Intelligence Ecosystem, designed to accelerate how new legislation gets drafted and updated as technology changes underneath it. And on the sidelines of the dialogue, Al Olama met directly with Guterres to reaffirm UAE support for a global AI cooperation framework, tying the country's domestic agentic rollout to the same international process Saudi Arabia was building institutional ties to a day earlier. Two Gulf states, two different entry points into the same conversation.

The 50 percent target deserves a second look, because it is not a marketing figure attached to a pilot program. It applies across ministries that handle payroll, permits, healthcare records and immigration, meaning the UAE is proposing to let autonomous systems make or shape decisions that already touch millions of residents. Al Olama's argument in Geneva was essentially that this scale is the point, not a risk to be managed around it. A government that runs agentic AI at that size, he suggested, generates exactly the kind of operational evidence the scientific panel says regulators elsewhere are missing.

UAE Government Agentic AI Adoption Target

Source: UAE government statement via Omar Sultan Al Olama, UN Global Dialogue on AI Governance, July 9, 2026

Doha and the Rest of the Room

Saudi Arabia and the UAE drew most of the regional attention, but they were not the only Gulf voices in Geneva. Qatar's minister of communications held meetings with heads of delegations and international organizations on the sidelines of the dialogue, a lower-profile role than Riyadh or Abu Dhabi's, though one that fits Doha's pattern of engaging multilateral tech forums without the infrastructure spending its neighbors have committed to. Bahrain and Kuwait were represented in the wider dialogue as UN member states, though neither has announced anything close to the UAE's agentic government targets or Saudi Arabia's UNESCO-linked research center.

That unevenness is worth naming plainly. The Gulf is not showing up in Geneva as a bloc with one strategy. It is showing up as two states with serious compute and institutional ambitions, and several more that are participating in the conversation without yet having the infrastructure to back a seat at its center. For companies and investors watching the region, the distinction matters more than the Gulf Cooperation Council label suggests. Governance leverage in AI is not being distributed evenly even within the six member states.

Building Fast and Writing Rules at the Same Time

There is an obvious tension in a government importing 35,000 of the most advanced chips available while also sending a delegation to a summit about the danger of AI moving faster than oversight can follow. Saudi and Emirati officials would argue there is no contradiction at all, because the two moves solve different problems. The chip deal buys sovereignty over where AI training happens. The Geneva diplomacy buys influence over how the rest of the world writes rules for AI everywhere else, including systems trained outside Gulf borders that Gulf citizens and companies will still use every day.

Whether that logic holds depends on questions Geneva did not answer. The scientific panel's warning about an evidence dilemma cuts against any government racing to deploy first and regulate later, and both Gulf states are doing exactly that with their domestic agentic rollouts. Saudi Arabia's own SDAIA ethics principles and the UAE's Regulatory Intelligence Ecosystem are real institutional commitments, not just talking points, but neither has been tested against an incident serious enough to prove whether the guardrails hold under pressure. The forty scientists who wrote the July 1 report were careful to say the same thing about every government at the table, not only the ones in the Gulf. None of the 40-plus governance frameworks currently in existence, anywhere, has been stress-tested against a real crisis.

There is also a quieter calculation at work. Every month the Gulf spends building compute and institutions is a month the region is not waiting for Brussels or Washington to define what compliant AI looks like for a bank in Riyadh or a hospital in Abu Dhabi. Owning the infrastructure means owning the timeline. If a binding global framework eventually does emerge from a future round of this same UN dialogue, Gulf officials will be negotiating from inside the room that built the report, not reading a document handed to them after the fact.

What It Means for the App on Your Phone

Summit diplomacy can feel distant from a product roadmap, but the compliance environment taking shape around it is not. A developer building an agentic banking assistant or a delivery app with autonomous ordering in Riyadh or Dubai today is already answerable to SDAIA's ethics principles and the emerging UAE regulatory ecosystem, frameworks that were written domestically and now carry the added weight of having been presented to the UN as models worth exporting. Neither framework yet spells out granular technical requirements for consumer apps the way the EU's AI Act does, but both signal where enforcement priorities will land first: financial services, government-adjacent services, and anything processing citizen data at scale.

The practical takeaway for founders and product teams in the region is to stop treating AI governance as a Brussels or Washington problem to monitor from a distance. Saudi Arabia's September ethics forum and the UAE's regulatory intelligence platform are both explicitly designed to produce rules faster than traditional legislative cycles allow, which means the compliance bar for agentic consumer products in the Gulf could move before global standards catch up. Teams shipping agentic features into GCC markets this year would be better served treating local ethics principles as binding now rather than waiting for a formal law to force the issue.

The Calendar That Actually Matters Next

Geneva was a dialogue, not a decision point, and both the UN's own framing and the Gulf's follow-through calendar reflect that. The next test comes in September, when Saudi Arabia hosts the fourth Global Forum on the Ethics of Artificial Intelligence in Riyadh from the 14th to the 17th, an event UNESCO organizes annually and one that will be watched for whether the kingdom's new research center produces anything beyond a ribbon-cutting. The UAE's two-year agentic government target gives it until roughly 2028 to prove Al Olama's Geneva framing right, that governing outcomes rather than individual models actually scales when half a country's bureaucracy runs on it.

For the UN process itself, the Independent Scientific Panel is expected to keep issuing updated assessments as the political dialogue continues into future sessions, meaning the 90 percent compute concentration figure and the evidence dilemma framing will likely resurface at the next major AI policy gathering rather than fade after one news cycle. Investors and operators tracking the Gulf's AI ambitions have a narrower set of dates to watch than the sprawling summit itself suggests: September in Riyadh, and whatever milestone the UAE publishes en route to its 2028 target.

The Analyst's Read

Strip out the ceremony and Geneva left behind two facts worth carrying forward. First, the scientific consensus on AI risk has moved from academic hedge to formal UN record, which raises the cost of any government, Gulf or otherwise, ignoring it entirely. Second, the Gulf states did not treat that record as a reason to pause. They treated it as a deadline to become credible participants in whatever comes after Geneva, and they used chips, institutions and September calendar dates to do it rather than statements alone.

Whether that strategy ages well depends on something no summit could settle: what happens the first time an agentic system deployed under one of these frameworks fails in public, in a hospital, a bank, or a government office running at the pace Abu Dhabi has committed to. Saudi Arabia and the UAE are betting they can build the plane and the maintenance manual at the same speed. Geneva's scientists spent two days explaining, politely, why that bet is not obviously safe. Neither side changed course because of it, and the next real evidence either way arrives in Riyadh in September.