The Sovereignty Question: Who Governs the Governors?

Contents

A morning in a Tashkent bank, mid-1990s. On the director’s desk lies a fax bearing the IMF logo. The package is familiar by then: price liberalisation, full currency convertibility, accelerated privatisation. The same recommendations have reached every post-Soviet capital, delivered with the same confidence and the same expectation of compliance. Beside the fax lies an internal letter from the Central Bank of Uzbekistan. Its logic runs in the opposite direction. A managed exchange rate. A strict credit schedule. Continued state support for priority sectors. Two documents on one desk, drafted by institutions that describe themselves as working toward the same end, disagreeing about nearly everything that matters.

The director reads both. He does not raise his voice. He does not argue with the fax. He says one sentence, and that sentence captures the period more precisely than any analytical paper that would later be written about it: “We will listen, but we will live by our own rules.”

The scene is worth recording without ornament. No heroism. No retrospective claim that the decision was correct. The sentence was not a programme. It was a description of what was already happening. A sovereign state in the middle of a transformation was receiving a framework drafted elsewhere, acknowledging its receipt, and declining to be shaped by it. Thirty years later, in a completely different technological domain, the same mechanism is working again, under different names, with different actors, at a pace that no institution present in that room would have recognised.

Every enforcement architecture in AI governance meets one structural limit. The limit is the sovereign will of a state for which the technology has become an element of strategic autonomy. It is neither normative nor technical. It has reproduced in every governance architecture that has tried to constrain a sovereign interest, and in the AI domain it compresses into a window shorter than any prior sovereign conflict.

I. The Sovereign Override in Real Time

In early 2026, the dispute between Anthropic and the United States Department of Defense — which the Trump administration has recently rebranded with the secondary title “Department of War” [1] — enters open conflict. The contract under negotiation is worth approximately two hundred million dollars. The Department insists that its instance of Claude be usable for all lawful purposes, a phrase that, within the scope of defence work, includes autonomous targeting and domestic surveillance. Anthropic seeks contractual language that excludes specific categories of use. The Department declines the exclusions. It then designates Anthropic as a supply-chain risk, a category that carries procurement consequences beyond the single contract in dispute.

On 9 March 2026, Anthropic files suit in a federal court in the Northern District of California, challenging the designation [2]. The court issues a temporary pause. The case continues. In early April, press reports describe the British government approaching Anthropic with a view to expanding the company’s operations in the United Kingdom in the wake of the American conflict [3].

None of this is yet resolved. The point is not the outcome. The point is the mechanism visible in the sequence. A private actor declared a safety boundary. A sovereign actor with operational dependence on the same actor contested the boundary. The contest moved into court. A second sovereign, observing the contest, opened a parallel channel. At no point in the sequence did any multilateral body, professional standard, or industry norm function as a binding constraint. The only instruments in play were contract law, regulatory designation, litigation, and the competing interests of two states.

The essay makes no judgement on the parties. It records what happens to the structural integrity of a safety architecture when a sovereign actor has a strategic interest in shifting it. The shift is not settled. The mechanism is.

This picture is not an exception, and it is not an artefact of one contract or one company. Its reproducibility becomes visible only when one steps back and looks at the historical record of how enforcement architectures have actually lived with sovereign interests over the past half century.

II. The Historical Pattern

The Treaty on the Non-Proliferation of Nuclear Weapons

The Treaty came into force in 1970. India, Israel, and Pakistan never joined. North Korea joined and withdrew, with its notice delivered in January 2003 and the withdrawal taking effect that April [4]. Each of these four cases has persisted for decades. None has been resolved into treaty compliance. The International Atomic Energy Agency continues its verification work, the Treaty continues to function, and the documented exceptions continue to exist alongside it. The architecture does not close its exceptions when the sovereign interest of the excluded state is judged sufficiently strong and the cost of enforcement disproportionately high. This is not a failure of the architecture. It is its working state.

Financial sanctions and SWIFT

Financial sanctions and the SWIFT messaging system show a second form of the same mechanism. After 2014, and especially after 2022, Russia developed its System for Transfer of Financial Messages, expanded the Mir card network, and built operational links to China’s Cross-Border Interbank Payment System [5]. These alternative payment and messaging systems did not replace SWIFT. They reduced the cost of operating outside it. The effectiveness of SWIFT-based enforcement has depended throughout on the ratio between the cost of compliance, the cost of non-compliance, and the willingness of third-country institutions to accept the risk of secondary measures. The architecture operates selectively, by the logic of cost ratios rather than the logic of formal rules. Its effectiveness depends on political economy, not on the quality of drafting.

The IMF and Uzbekistan in the 1990s

The third precedent I know from inside the room. In the mid-1990s I was working in the banking system of Uzbekistan, in departments that received IMF correspondence directly. The standard package arrived: liberalisation of prices, full convertibility of the national currency, accelerated privatisation. Russia and Kyrgyzstan had accepted comparable packages earlier in the decade. The results were documented at the time and have been reconstructed many times since. In Russia, real GDP fell by approximately forty to forty-five per cent between 1991 and 1998, depending on the methodology of reconstruction. Hyperinflation in 1992 exceeded two thousand five hundred per cent. Excess mortality and declines in life expectancy through the transformational recession are documented in the demographic literature [6]. Kyrgyzstan, according to IMF Article IV consultations and World Bank transition surveys, recorded severe contraction and macroeconomic volatility in the same period, with heavy dependence on external financing [7].

Uzbekistan chose a different path. A managed exchange rate. Phased liberalisation. Continued state support for priority sectors. By the mid-1990s inflation had stabilised. Cumulative decline in GDP through 1990–1995 was approximately seventeen per cent, the smallest contraction among former Soviet republics over that period [8]. The choice was not a rejection of reform. It was a priority of governability over speed. And it worked only under two conditions that are easy to overlook: the state had its own interpretation of what it was doing, and it had personnel capable of implementing that interpretation from inside the institutions rather than translating a framework drafted elsewhere. Without those two conditions the same choice would have produced different results.

III. Why AI Compresses the Window

Three precedents. Three different correction horizons. The Treaty exceptions have persisted for more than fifty years. Sanctions adjustments run in years. The Uzbekistan choice of the 1990s unfolded across a decade. The AI domain now operates at a pace in which the same mechanisms complete their cycles in weeks and days. Three recent cases make this compression visible.

Level one — the Strait of Hormuz

In late February 2026, the United States and Israel struck Iranian territory in a coordinated campaign. On 4 March 2026, Iran declared the Strait of Hormuz closed and began attacks on vessels attempting to transit [9]. Daily vessel counts fell from around one hundred and thirty-five to fewer than ten within days. The International Energy Agency described the disruption as the largest supply shock in the history of the global oil market [10].

The blockade was aimed at energy flows. It was a strategic lever against the states that had struck Iran. But the same strait carries helium from Qatar and neon, argon, xenon, and krypton from several Gulf producers. These specialty gases are essential inputs for lithography and etching processes in semiconductor manufacturing. Without them, chip fabrication does not function. Approximately one third of global helium supply was taken out of circulation [11]. The disruption was not the declared aim of the blockade. It became its structural consequence.

The material limit of AI is activated by decisions taken for reasons unrelated to AI. It lies outside the jurisdiction of any safety architecture designed around AI. No framework drafted in San Francisco or Brussels could have reached into a sea lane in the Persian Gulf to prevent it.

Level two — the Chinese licensing regime

On 9 October 2025, the Ministry of Commerce of the People’s Republic of China announced a sweeping expansion of export controls over rare earth elements, related equipment, and processing technologies [12]. Five additional elements were added to the controlled list. For the first time, the scope extended extraterritorially to foreign-made products containing Chinese-origin materials or produced using Chinese technologies. In late October, at a meeting between Xi Jinping and Donald Trump on the sidelines of the APEC summit in Busan, a bilateral arrangement was reached. On 7 November 2025, China formally suspended the October expansion for one year, until 10 November 2026, in exchange for a one-year suspension of the United States BIS Affiliates Rule [13]. The April 2025 baseline controls remained in force.

The full cycle — announcement, material response across global supply chains, political pressure, bilateral arrangement, partial suspension — took approximately one month. At no point in the cycle did an institutional enforcement architecture operate. The correction instrument was a bilateral deal between two states. The licensing regime itself is discretionary: not a binary prohibition, but a continuing capacity to grant or withhold permission on any application, at any time, on any ground.

Level three — Resolution No. 109

On 18 March 2026, the Cabinet of Ministers of Uzbekistan adopted Resolution No. 109 on the regulation of paper use in the digitalisation of executive agencies [14]. All republican and local bodies are required to transition to paperless document flow by 2030. Procurement of paper, ink, and printers is reduced by fifteen per cent annually. A national paperless day is established. Waste-paper collection containers are installed in state offices. The ratio of one printer per twenty staff is set. Compliance is integrated into agency performance evaluation systems. The only exception is documents classified as state secrets.

The text of the Resolution contains no fallback procedures, no plan for return to paper-based operation, and no requirement to maintain parallel competencies. This is a documented absence, not an interpretation. On the side of the state receiving the technology, the same mechanism operates through a sovereign instrument that does not include a correction-window category because that category is not part of the institutional planning grammar of the receiving side.

If a sovereign gesture in one point of the planet immediately creates a material limit in another, and if the only instrument for its correction is a bilateral arrangement between states, what remains at the disposal of any safety framework that attempts to measure not intent but reality?

IV. The Structural Implication

The question changes. The framing that asks who should govern the governors is a normative question. It does not have an analytical answer. It leads to proposals about which institution, which treaty, which charter should be written by whom, with what authority. None of the prior three precedents was closed by answering that question. Each continues to operate with the sovereign limit built into it as a permanent feature.

The analytical question is different. It asks what structurally happens when sovereign interest contradicts a safety framework in a domain with a short correction window. The answer does not prescribe. It describes.

An enforcement architecture that does not account for sovereign override measures its own intent. It does not measure reality. It continues to measure its own intent until the window closes. The Treaty architecture measured its intent across decades while exceptions persisted outside it. The sanctions architecture measured its intent across years while alternative systems were built alongside it. The Uzbekistan choice of the 1990s played out across a decade of divergent outcomes. The AI domain now operates at a pace at which these periods collapse into a single interval too short for any enforcement coalition to form, negotiate, ratify, and implement.

Three features of the domain produce this compression together. The first is the speed of deployment, which runs on release cycles rather than on diplomatic calendars. The second is the general-purpose nature of the technology, which makes it strategically valuable across defence, commerce, administration, and research at the same time. The third is the agency transfer analysed in the previous essay: as decisions migrate from human operators to systems, the competence required to operate a fallback architecture atrophies in the same institutions that would need to deploy it [15].

These three features together compress the correction window below the interval in which institutional adaptation has historically taken place. This is the state of the domain, not a prediction about it. The state is visible in the three cases already described, each of which completed its cycle within a timeframe that the Treaty, the sanctions regime, and the Uzbekistan transition could not have accommodated.

Implications

First: every enforcement architecture designed without a model of sovereign override will report on its own intent rather than on its effect. This reporting will continue until the correction window closes.

Second: the correction instruments available in the AI domain are contract, regulatory designation, litigation, and bilateral arrangement. Multilateral instruments operate on horizons longer than the domain allows.

Third: material limits activated by decisions unrelated to AI cannot be corrected by any framework whose scope is defined through AI. The jurisdiction gap is structural, not legal.

Fourth: on the side of states receiving the technology, institutional planning grammar currently lacks a correction-window category. Sovereign decisions are formalised through instruments that treat the transition as one-directional.

Signals to Watch

  • Defence contracts held by major AI laboratories and the contractual language used to describe permitted use.

  • Export controls on compute hardware and their renegotiation through bilateral channels.

  • National AI strategies published by states that do not produce frontier models, and the absence or presence of correction-window provisions within them.

  • Licensing activity under the Chinese rare earth controls after 10 November 2026, when the one-year suspension expires.

  • Global semiconductor supply chain responses to any further escalation in the Persian Gulf or in other maritime chokepoints.

The Questions That Remain Open

The sovereign override explains why an enforcement architecture does not hold its boundary under political pressure from a state with a strategic interest in shifting it. It does not explain a different limit. The different limit operates through matter rather than through will. It is not overcome by any sovereign instrument because it does not answer to any sovereign instrument.

The declaration of sovereignty and the material capacity to exercise it are two different things. In the AI domain, the gap between them is structural. A state can issue a resolution. It cannot issue a fabrication line. It can license an export. It cannot manufacture the physical substrate from which the export is composed. The next essay in this series examines that second limit on its own terms.

Two questions remain open at the boundary of this essay and the next. Can any enforcement architecture accommodate both limits at once — the limit of will and the limit of matter — within a correction window compressed below the interval of institutional adaptation? And if it cannot, what does the absence of such an architecture mean for the states that are neither producers of the technology nor sovereign drivers of its direction, but receivers of its consequences?

Sources & Notes

[1] The White House. Executive Order 14347 — Restoring the United States Department of War. 5 September 2025. The order authorises the Department of Defense to use “Department of War” as a secondary title in non-statutory communications. The legal name remains Department of Defense; only an Act of Congress can formally rename a federal department. whitehouse.gov

[2] Reuters. “Anthropic sues to block Pentagon blacklisting over AI use restrictions.” 9 March 2026. reuters.com. See also Essay 6 of this series: okhodjaev.com/essays/the-pattern-closes/ and Essay 8: okhodjaev.com/essays/the-agency-transfer/

[3] Reuters. “Britain woos Anthropic expansion after US defence clash, FT says.” 5 April 2026. reuters.com

[4] International Atomic Energy Agency. Treaty on the Non-Proliferation of Nuclear Weapons — Status of Adherence. Updated 2025. iaea.org See also: United Nations Office for Disarmament Affairs. Treaty on the Non-Proliferation of Nuclear Weapons (NPT). un.org

[5] Bank for International Settlements. Cross-border payments — work programme. 2024. bis.org See also: Atlantic Council GeoEconomics Center. Russia, China, and the Future of the Dollar. 2023.

[6] Aslund, A. Russia’s Capitalist Revolution: Why Market Reform Succeeded and Democracy Failed. Peterson Institute for International Economics, 2007. See also: Popov, V. “Shock Therapy versus Gradualism Reconsidered.” Comparative Economic Studies, 2007. Demographic data: Brainerd, E. “Reassessing the Standard of Living in the Soviet Union.” Journal of Economic History, 2010.

[7] International Monetary Fund. Kyrgyz Republic — Article IV Consultation Reports, 1995–1999. imf.org See also: World Bank. Transition: The First Ten Years — Analysis and Lessons for Eastern Europe and the Former Soviet Union. 2002. worldbank.org

[8] Pomfret, R. The Central Asian Economies in the Twenty-First Century: Paving a New Silk Road. Princeton University Press, 2019. See also: International Monetary Fund. Republic of Uzbekistan — Recent Economic Developments, IMF Staff Country Report No. 98/116, 1998. imf.org

[9] NPR. “Iran effectively closes Strait of Hormuz as U.S.-Israel strikes continue.” 12 March 2026. npr.org See also: Congressional Research Service. Iran Conflict and the Strait of Hormuz: Impacts on Oil, Gas, and Other Commodities. March 2026. congress.gov

[10] International Energy Agency. Strait of Hormuz Factsheet. February–March 2026. iea.org

[11] World Economic Forum. “Beyond oil: 9 commodities impacted by the Strait of Hormuz crisis.” April 2026. weforum.org See also: Table.media. When the Strait Closes: Trade Dependencies and Shipping. Research brief, March 2026.

[12] Center for Security and Emerging Technology, Georgetown University. Ministry of Commerce Notice 2025 No. 61: Announcement of the Decision to Implement Controls on Exports of Rare Earth-Related Items. Translation and analysis. November 2025. cset.georgetown.edu See also: White & Case LLP. “China imposes extraterritorial jurisdiction and a 50% Rule for export controls on rare earth elements and other items.” 13 October 2025. whitecase.com

[13] Dezan Shira & Associates. China Briefing — China’s Rare Earth Export Controls: Impact on Businesses and Industries. Updated November 2025. china-briefing.com See also: Center for Strategic and International Studies. “China’s New Rare Earth and Magnet Restrictions Threaten U.S. Defense Supply Chains.” October 2025. csis.org

[14] Cabinet of Ministers of the Republic of Uzbekistan. Resolution No. 109 of 18 March 2026 — On measures to regulate the use of paper within the framework of digitalization of work processes in the activities of republican and local executive authorities. lex.uz/docs/8090805 See also reporting in UzDaily and Gazeta.uz, March 2026.

[15] See Essay 8 of this series: okhodjaev.com/essays/the-agency-transfer/

Full essay and updated sources: okhodjaev.com/essays/the-sovereignty-question/

Oybek Khodjaev: systems transformation analyst, Founder & CEO of INVEXI LLC. Former Deputy Governor (Deputy Khokim) of Samarkand Region. Previously, Treasury Director and Deputy Chairman of the Management Board at JSC UzAgroIndustrialBank. More than thirty years’ experience in economics, banking, finance, and business across Uzbekistan and the CIS.

Published April 13, 2026 on https://okhodjaev.com/essays/the-sovereignty-question/

Substack · X.com · LinkedIn

Published