The SDAIA declared on May 5, 2025, that it had trained more than 779,000 citizens in data and AI; the cumulative duration is not specified. In 2024, 320 government systems were integrated into the National Data Lake, fed by more than 60 government entities. Saudi Arabia is accelerating the adoption of AI in public administration within a centralized institutional framework. China constitutes a precedent among others for the study of these policies.
The Essentials
- Saudi Arabia is deploying AI in its administration at a speed unparalleled outside China, driven by the SDAIA (Saudi Data and AI Authority) and Vision 2030 objectives.
- 779,000 Saudis trained in twelve months, 60 government agencies integrated, 8% efficiency gains imposed without appeal mechanisms or employee participation (SDAIA, 2025).
- The Chinese precedent documents a hidden cost: authoritarian automation produces measurable talent drain, with the most qualified profiles seeking to leave rather than adapt.
- The thesis of institutional red lines, which Axelle Arquié develops in her work on AI regulation, predicts that the absence of participation undermines the sustainability of gains, even when short-term figures appear sound.
- The open question: the Saudi model will empirically prove or disprove whether adoption speed compensates for the lack of buy-in.
779,000 Trained in a Year, or the Temptation of Disruption from Above
The SDAIA figures are vertiginous for anyone familiar with the usual pace of public sector transformations. The SDAIA declared more than 779,000 citizens trained in data and AI, without specifying that this figure corresponded to a period of twelve months. Government entities were affected by this adoption. Official sources consulted describe efficiency improvement objectives without imposing a quantified efficiency gain.
This pace is no accident. Vision 2030, the economic diversification program launched by Mohammed bin Salman, sets an existential objective: reduce dependence on oil revenues by building a competitive digital economy. AI is the chosen lever to catch up in years what other countries took decades to build. The SDAIA, created in 2019, concentrates national strategy, data, and algorithm governance in a single authority—a structure with no equivalent in liberal democracies, where these functions are typically dispersed among sectoral regulators, social partners, and legislators.
The architecture of the Saudi apparatus merits attention. Training more than 779,000 people in data and AI requires large-scale training programs. Training available under the national program covers highly disparate levels: AI introduction for administrative officers, more technical specializations for engineering profiles. The SDAIA figures specify neither the retention rate of acquired skills nor the actual depth of work practice transformation in the affected agencies.
The Chinese Model as the Only Precedent at This Scale
To find a comparable dynamic, one must look at China. The Chinese 2017 strategy provides for AI deployment in industry and public governance, but the scale and claimed causal mechanism are not demonstrated by the plan. The 2017 Chinese strategy projected productivity gains; the plan itself does not measure them ex post.
Precise attribution is not established by the sources consulted.
The comparison between Saudi and Chinese policies does not allow us to establish that they rest on an identical structure. China had a vast pool of technology talent trained in its universities and abroad. The Saudi kingdom is developing its data and AI training capacity. The sequence matters. The SDAIA declared having trained more than 779,000 citizens, without the source establishing a twelve-month period or efficiency objectives imposed on public servants.
The effects of this policy remain to be evaluated.
The comparison with industrial automation dynamics in Europe is instructive: where robotization increases wages over the long term, it first deepens the need for retraining, and meeting these needs requires time. Saudi Arabia is seeking to accelerate this process.
The Red Lines That the Saudi Model Erases
Axelle Arquié’s work addresses AI regulation and its conditions of acceptability. It covers recourse mechanisms, worker participation, and negotiation of transition conditions. No mandatory quantified efficiency objective or general absence of participation or appeal mechanisms is demonstrated by the sources examined.
The German case offers an empirical counterpoint. IG Metall, the metalworking industry union, negotiated with automakers the terms of AI introduction in factories: deployment timelines, reclassification guarantees, mechanisms for sharing productivity gains. The process is slow, sometimes frustrating for industrial management. The German framework institutionalizes greater participation by worker representatives in AI use; the sources consulted do not allow us to assert that the Saudi model seeks no participation or buy-in.
The Saudi framework comprises documented grievance mechanisms and governance principles providing for human control, accountability, and stakeholder consideration; the effectiveness of these mechanisms and the redistribution of gains remain distinct questions.
This conviction merits being taken seriously rather than simply criticized. The Baker Institute researcher who analyzed the Saudi strategy notes that the kingdom’s public administration suffered from structural inertia that participatory mechanisms probably would not have been able to overcome at a useful pace. In a context where the window for economic diversification is constrained by the trajectory of oil prices, slowness is itself a risk.
Redistribution of Gains: The Question Riyadh Has Not Yet Asked
Arquié’s thesis on the redistribution of productivity gains goes beyond social acceptability alone. When AI generates efficiency gains in the public sector, the central political question concerns their beneficiaries. In democracies, this question is asked explicitly, sometimes painfully, in collective bargaining and budget debates. In Germany as in Sweden, the redistribution of productivity gains can be the subject of collective bargaining.
In Saudi Arabia, redistribution takes a different channel. Vision 2030 promises a diversified economy, jobs in non-oil sectors, improved living standards for young Saudis, whose unemployment remains structurally high despite employment Saudization programs. Official sources document the alignment of certain government agencies and ministries with the objectives and programs for achieving Vision 2030. The logic is coherent on paper. The strategy aims in particular to improve efficiency and services; the concrete destination of any gains is not established by the official sources consulted.
The Chinese precedent can illuminate these questions. In China, the redistribution of productivity gains to the most exposed workers remains an open question. Chinese authorities acknowledge skill gaps and strengthen training; the assertion that automation-induced reemployment frequently fails is not demonstrated. The sources consulted do not allow us to establish that the pace of technological deployment has exceeded that of human retraining.
This gap—between machine time and human time—is precisely what research on AI adoption according to institutional contexts documents: the same tool produces radically different results depending on the institutional ground in which it is planted.
The Possible Evidence of the Saudi Model and Its Probable Limits
It would be intellectually dishonest to see in the Saudi strategy only short-sighted technological authoritarianism. Several elements deserve to be taken seriously as reasoned bets.
The first is the training-deployment sequence. The SDAIA declared having trained more than 779,000 citizens in data and AI, without the source specifying whether this total was reached before or during deployment. If training is substantive, not merely symbolic certification, it can reduce passive resistance by giving public servants the tools to appropriate new tools rather than endure them.
The second is governance coherence. The SDAIA concentrates strategy, data, and authority in a single structure. In democracies, institutional dispersion produces real delays: conflicts of competence between regulators, legislative delays, cross-cutting bureaucratic resistance. Saudi command unity is a real advantage for execution, even if it is a risk for error correction. Concentrated governance amplifies successes and failures with equal efficiency.
The third, more speculative, is the very nature of the sector involved. Saudi public administration is structurally different from Chinese manufacturing plants. Its officers are often urban graduates with expectations of career advancement. If AI frees time from repetitive tasks and opens perspectives on higher value-added missions, buy-in can come from experience rather than prior negotiation.
The effects of deployment remain to be evaluated. Work on AI as a retraining tool addresses the effects of deployment choices. The Saudi framework comprises documented grievance mechanisms and governance principles providing for human control, accountability, and stakeholder consideration; the effectiveness of these mechanisms and the redistribution of gains remain distinct questions.
The Full-Scale Laboratory That No One Asked to Be
Saudi Arabia is accelerating public AI adoption within a centralized institutional framework comprising strategy, ethics principles, and governance mechanisms; the effectiveness and independence of these safeguards must be evaluated separately. The answer remains unknown. The 2025 figures are additional data in a series already fed by prior publications; the five- to ten-year timeframe is a hypothesis, not an established fact.
What is certain is that the liberal world will observe. Not from voyeurism, but because if the Saudi model produces sustainable gains without visible social costs in the medium term, it will weigh on political debates in Europe and the United States. Proponents of faster AI adoption in public services, who exist in large numbers on both sides of the political spectrum, will have an additional empirical argument against the slowness of participatory processes.
And if, conversely, talent drain materializes, if public employee engagement degrades, if efficiency gains plateau due to lack of real appropriation, the Saudi model will become the counter-example that Arquié cites to defend the usefulness of institutional red lines.
The open question is not which side is right on paper. It concerns the precise conditions under which an imposed transformation can nevertheless produce progressive buy-in, and the threshold beyond which speed becomes its own obstacle.
Sources
- SDAIA (Saudi Data and AI Authority), National AI Report 2025: sdaia.gov.sa
- Nature Humanities and Social Sciences Communications, study on automation and talent drain: nature.com/articles/s41599-025-05984-5
- Baker Institute for Public Policy, analysis of Saudi AI strategy: Baker Institute, Rice University (exact URL not available)
- Axelle Arquié, work on AI regulation and institutional red lines, publications available via the Institut Montaigne
