Britain’s struggle to govern frontier artificial intelligence exposes a fundamental transformation in state power. Governments retain the authority to make laws, but some of the capabilities most consequential to national security are developed, operated and controlled by private technology companies beyond their direct command.
By Vladimir Tsakanyan, Ph.D.
Center for Cyber Diplomacy and International Security (CCD-IS)
The State Confronts a New Kind of Power
For centuries, sovereignty has been associated with the ability of governments to exercise authority within their territory.
States make laws, enforce regulations, maintain armed forces, protect borders and determine the institutional rules under which economic activity operates.
Private companies can accumulate enormous wealth and influence, but they remain subject to public authority.
Artificial intelligence is complicating that traditional relationship.
On October 17, 2026, growing concern within Britain’s governing Labour Party brought renewed attention to the speed of frontier-AI development and the government’s ability to manage its risks.
The immediate debate concerns cybersecurity, biological threats, autonomous systems, public safety and the adequacy of existing regulatory arrangements.
But beneath these concerns lies a deeper constitutional and geopolitical question.
What happens when a government becomes dependent on technological capabilities that it neither owns nor fully controls?
Britain provides an unusually revealing example.
It has established one of the world’s leading public institutions for evaluating advanced AI systems.
Its government increasingly uses frontier models to strengthen cybersecurity.
Its intelligence and security institutions recognize AI as a rapidly developing strategic capability.
Yet Britain does not possess a domestic frontier-model industry comparable in scale to the largest American developers.
Its independent testing capability therefore depends substantially on access negotiated with private laboratories and international partners.
This is the sovereignty paradox.
The government is responsible for protecting the public from the risks of advanced AI.
But its ability to understand and manage those risks depends partly on cooperation from the organizations creating them.
The Historical Foundations of Technological Sovereignty
The relationship between technological capability and state power has changed repeatedly throughout modern history.
During the nineteenth century, industrial capacity became central to national power.
Railways, steel production, telecommunications and advanced manufacturing shaped economic competitiveness and military capability.
Private corporations built much of this infrastructure.
Yet governments developed regulatory systems and strategic relationships to ensure that essential national interests were protected.
Railways could be privately operated, but governments regulated their safety and strategic use.
Telecommunications companies could own infrastructure, but states established legal authority over communications and national-security access.
Electricity generation increasingly involved private investment alongside public regulation.
The principle was that privately operated infrastructure could remain subject to public rules.
Technological sovereignty did not require governments to own every factory, railway or communications network.
It required them to maintain sufficient authority and institutional capacity to protect public interests.
Artificial intelligence tests whether this historical model remains adequate.
The Nuclear Precedent: When Strategic Technology Became State Power
The development of nuclear technology during the Second World War offers an important comparison.
The Manhattan Project demonstrated that scientific research could produce capabilities with consequences extending far beyond commercial or academic activity.
The United States government mobilized universities, laboratories, industrial companies and military institutions to develop nuclear weapons.
Private organizations contributed essential expertise.
But the program operated under centralized state direction.
After the war, governments established specialized institutions and legal frameworks to regulate nuclear technology.
The United States created the Atomic Energy Commission in 1946.
International institutions subsequently emerged to address nuclear safety, safeguards and proliferation.
The International Atomic Energy Agency was established in 1957.
The nuclear example demonstrates how governments historically responded when a technology acquired extraordinary strategic importance.
They developed public institutions capable of supervising, controlling and assessing it.
Artificial intelligence differs fundamentally from nuclear technology.
AI is commercially distributed, highly general-purpose and embedded across civilian economic activity.
Its capabilities are not confined to a narrow category of physical materials or facilities.
Nevertheless, the historical comparison raises a useful question.
If frontier AI increasingly possesses national-security significance, can governments rely primarily on voluntary cooperation with its developers?
The Telecommunications Precedent
Telecommunications offers another instructive comparison.
During the twentieth century, communications infrastructure became essential to national administration, economic activity and military coordination.
Many governments initially treated telecommunications as a public utility or state monopoly.
Others relied on regulated private operators.
In the United States, AT&T occupied a uniquely important position within the national communications system.
In Britain, telecommunications infrastructure was historically associated with the state-operated Post Office and later British Telecom.
The privatization of British Telecom in 1984 did not eliminate the government’s regulatory responsibilities.
Instead, it changed the institutional mechanisms through which public authority was exercised.
Governments increasingly relied on licensing, technical standards, competition rules and security obligations.
The lesson is significant.
Private ownership and national sovereignty are not inherently incompatible.
But sovereignty requires more than legal declarations.
Regulators must possess the technical knowledge, enforcement authority and operational visibility necessary to oversee essential systems.
That requirement becomes difficult when the technology develops faster than the institutions supervising it.
The Digital Platform Era
The internet introduced a different model of power.
A relatively small number of technology companies came to operate platforms serving billions of users.
Search engines organized access to information.
Social networks influenced public communication.
Cloud providers supplied computing infrastructure.
Mobile operating systems became gateways to digital services.
Governments retained jurisdiction over companies operating within their territories.
But regulatory authority became more difficult to exercise.
Platforms operated across borders.
Data moved internationally.
Corporate technical decisions affected entire societies.
Governments frequently struggled to understand the systems they were attempting to regulate.
The resulting debates concerned privacy, competition, content moderation, taxation, national security and democratic accountability.
Artificial intelligence inherits many of these problems.
But it adds something important.
AI systems do not merely distribute information or provide computing infrastructure.
They increasingly perform cognitive and technical tasks.
They write software.
Analyze documents.
Assist scientific research.
Identify vulnerabilities.
Support intelligence analysis.
Generate recommendations.
And, in some applications, take actions through connected tools.
The private technology provider is therefore no longer simply operating infrastructure.
It may be supplying capabilities that directly affect how institutions make decisions and perform security functions.
Britain as a Case Study
The United Kingdom has pursued an approach emphasizing innovation, technical evaluation and cooperation with leading AI developers.
In November 2023, Britain hosted the AI Safety Summit at Bletchley Park.
The location was historically significant.
During the Second World War, Bletchley Park served as a center of British codebreaking and signals-intelligence work.
The summit brought governments, technology companies and researchers together to discuss the risks associated with advanced AI.
The Bletchley Declaration recognized the potential for serious harm arising from frontier AI and the need for international cooperation.
Britain also established the AI Safety Institute, subsequently renamed the AI Security Institute.
The institution was designed to develop independent technical understanding of advanced AI capabilities and risks.
This represented an important innovation in governance.
Rather than relying exclusively on legal rules or company statements, the government began building the expertise required to test advanced systems.
But the institutional design also exposed a limitation.
Testing requires access.
And access to privately developed frontier models is not automatically guaranteed by the existence of a government research institution.
The October 2026 Access Problem
On October 12, AI Minister Kanishka Narayan confirmed that Britain’s AI Security Institute was evaluating a new generation of frontier models.
The announcement followed cooperation with American authorities and major AI developers.
The government emphasized the importance of maintaining access to advanced systems before public release.
This is strategically important.
Early access allows independent researchers to identify potentially dangerous capabilities and assess safeguards before widespread deployment.
Without such access, government evaluations may occur only after a model has entered public use.
But the arrangement depends on cooperation.
Britain’s technical capacity to evaluate frontier models does not automatically provide an enforceable right to inspect every model under development.
The distinction between regulatory authority and technical access becomes visible.
A government can possess the authority to legislate.
It can establish a research institute.
It can employ technical experts.
Yet it may still need permission from a foreign company to examine a strategically important system.
This is a new form of technological dependence.
The Difference Between Testing and Regulation
The October parliamentary debate revealed another important distinction.
Technical evaluation is not the same as regulation.
An institute may discover a dangerous capability.
It may identify a weakness in a model’s safeguards.
It may advise government officials.
It may publish research.
But these functions do not necessarily include the legal authority to compel a developer to modify, delay or withdraw a system.
That requires a separate statutory or regulatory basis.
At a parliamentary hearing on October 13, representatives from OpenAI and Anthropic supported stronger legally binding AI regulation.
Representatives from Google DeepMind and Meta expressed different preferences, emphasizing existing regulatory structures and technical standards.
The disagreement illustrates the diversity of views within the technology industry.
It also reveals a broader governance problem.
When companies disagree over the appropriate limits of their own technology, who ultimately determines the public interest?
In a democratic political system, that responsibility belongs to legitimate public institutions.
But effective exercise of that authority requires technical competence and enforceable legal powers.
Voluntary Cooperation Is Not Public Authority
Voluntary agreements can be useful.
They allow governments to obtain information quickly.
They encourage technical collaboration.
They can support safety evaluations before legislation is completed.
They may be particularly valuable when technology changes rapidly.
But voluntary cooperation has structural limitations.
Participation can depend on corporate priorities.
Access may vary between companies.
Disclosure practices may differ.
A developer may change its internal policies.
Commercial competition may create incentives to accelerate deployment.
Governments may find themselves negotiating access rather than exercising an established inspection right.
This does not mean companies are necessarily acting irresponsibly.
It means that voluntary cooperation and public accountability perform different functions.
A government cannot permanently substitute one for the other.
The Strategic Dependency Problem
The sovereignty paradox becomes more serious when governments begin using frontier AI for their own security functions.
Britain’s October 12 announcement described government programs using advanced AI to identify cybersecurity weaknesses.
Officials reported that pilot initiatives had helped identify and address 1,352 vulnerabilities across 548 code repositories in 18 departments.
The government announced an expansion of this work.
The operational benefit is substantial.
AI can help security teams examine software, identify vulnerabilities and improve defensive efficiency.
But reliance on frontier models also creates new questions.
Who controls the model?
Where is it operated?
What happens when its behavior changes?
Can the government audit its outputs?
What happens if access is interrupted?
Can sensitive government information be processed without unacceptable exposure?
Does the government possess alternatives?
The more strategically useful a privately controlled model becomes, the more important these questions become.
Sovereignty Without Ownership
It would be a mistake to conclude that governments must own every advanced AI system.
Modern states depend on private industry in numerous strategic sectors.
Defense contractors manufacture military equipment.
Private companies operate telecommunications networks.
Commercial satellite providers supply important capabilities.
Cloud companies support government computing.
The central issue is not ownership alone.
It is control over essential conditions.
A sovereign government should be able to determine the legal rules governing technology deployed within its jurisdiction.
It should possess sufficient technical expertise to evaluate risks.
It should be able to enforce legitimate requirements.
It should preserve operational continuity when suppliers fail.
And it should avoid dependencies that make essential public functions vulnerable to unilateral commercial decisions.
Technological sovereignty therefore concerns institutional capacity rather than simple nationalization.
The Three Dimensions of AI Sovereignty
AI sovereignty can be understood through three interconnected dimensions.
The first is legal sovereignty.
This concerns the authority to establish and enforce rules governing AI development, deployment and use.
The second is technical sovereignty.
This concerns the ability to understand, evaluate, audit and, where necessary, independently operate relevant AI capabilities.
The third is operational sovereignty.
This concerns the ability to maintain essential public services and security functions despite disruptions involving particular vendors, models or foreign infrastructure.
A country may possess one dimension without the others.
It may regulate AI while lacking independent testing capabilities.
It may operate domestic computing infrastructure while depending on foreign models.
It may possess advanced research institutions while lacking enforceable oversight powers.
The sovereignty paradox emerges when governments assume that legal authority automatically provides technical and operational control.
It does not.
The International Dimension
Britain’s position is not unique.
Many countries are developing national AI strategies while relying on technologies produced by a relatively small number of companies.
The concentration of frontier-model development creates international dependencies.
Governments may depend on foreign cloud infrastructure.
They may rely on imported semiconductors.
They may use externally developed models.
They may lack domestic expertise for evaluating advanced capabilities.
These dependencies are particularly significant for smaller and middle powers.
A country may possess strong democratic institutions and advanced regulatory systems while remaining technologically dependent on suppliers outside its jurisdiction.
This creates an asymmetry between political sovereignty and technological capability.
The international AI order may therefore increasingly distinguish between countries that develop frontier systems and countries that primarily consume or regulate them.
The United States and the Politics of Innovation
The United States possesses significant advantages in frontier AI.
Major American companies develop some of the world’s most capable models.
The country benefits from advanced research institutions, private investment, cloud infrastructure and a large technology ecosystem.
American policymakers also face competing priorities.
They seek to preserve technological leadership.
They want companies to innovate.
They recognize national-security risks.
They must consider public safety and economic competition.
The resulting policy debate often centers on the relationship between innovation and regulation.
Some policymakers argue that excessive restrictions could weaken American competitiveness, particularly relative to China.
Others contend that insufficient oversight could create security and public-safety risks.
These positions reflect different assessments of how technological leadership should be protected.
The sovereignty question becomes more complicated for American allies.
Their security may depend on access to American technology.
But their regulatory preferences may differ from those of Washington.
Europe’s Alternative
The European Union has pursued a more explicitly legislative approach.
Its AI Act establishes a risk-based regulatory framework covering different categories of AI systems, including obligations concerning general-purpose AI.
The legislation entered into force in 2024, with provisions applying according to a phased timetable and subsequent legislative developments.
Its implementation should not be described as a single completed event.
The European approach demonstrates an attempt to exercise regulatory sovereignty over AI systems operating within the European market.
But legal regulation does not eliminate technological dependence.
Europe can establish requirements for AI providers.
It cannot automatically create domestic alternatives to every leading foreign model.
Nor can it guarantee that all necessary computing resources will be available within Europe.
The European challenge therefore involves both regulation and industrial capability.
Rules determine how technology may operate.
Investment determines whether alternatives can be developed.
Sovereignty requires attention to both.
China’s Different Model
China approaches AI through a different political and institutional structure.
The Chinese government has established regulations concerning algorithms, deep synthesis and generative AI services.
It also supports domestic technological development and seeks to reduce dependence on foreign strategic technologies.
This model gives the state extensive authority over domestic digital services.
But government control does not automatically resolve all problems of AI governance.
Strong state authority can coexist with technological uncertainty, economic incentives and risks associated with rapidly advancing systems.
The comparison illustrates an important distinction.
AI sovereignty is not synonymous with democratic accountability.
A government may possess extensive technical control while offering limited independent scrutiny.
Conversely, a democratic government may maintain stronger legal safeguards while struggling to secure technical access.
The quality of governance depends on both capability and accountability.
The Return of Industrial Policy
The growing importance of AI has revived interest in industrial policy.
Governments are investing in computing infrastructure, semiconductor supply chains, research institutions and domestic AI development.
The logic resembles earlier periods of strategic technological competition.
During the Cold War, states supported aerospace, electronics, nuclear research and advanced computing.
Today, AI infrastructure is becoming an object of comparable strategic attention.
But building domestic capability is expensive.
Frontier AI requires advanced chips, energy, data centers, specialized researchers and sustained investment.
Not every country can reproduce the entire technological ecosystem.
Consequently, sovereignty may increasingly depend on diversification and trusted partnerships rather than complete self-sufficiency.
The objective is not necessarily technological isolation.
It is reducing the risks created by excessive dependence.
The Problem of Democratic Accountability
AI governance also raises questions about political legitimacy.
Private companies are accountable through corporate governance, market incentives, contractual obligations and applicable law.
Governments are accountable through constitutional institutions, elections, courts and legislative oversight.
These systems are not equivalent.
A company may make a technically reasonable decision that has significant public consequences.
A government may disagree with that decision but lack the technical knowledge to evaluate it.
A regulatory agency may possess legal authority but insufficient resources to enforce it.
The result can be a gap between formal responsibility and practical control.
This is especially concerning when AI systems affect public administration, security decisions or essential services.
The public may hold elected officials responsible for outcomes produced through systems that government institutions cannot fully inspect.
That is the democratic dimension of the sovereignty paradox.
Autonomous AI Intensifies the Problem
The emergence of increasingly autonomous AI agents makes these issues more urgent.
An AI system that provides information presents one category of risk.
A system that independently uses tools, modifies software or interacts with external infrastructure presents another.
The British government has begun developing plans for responding to serious incidents involving autonomous AI agents.
As of October 17, these plans remain under development.
They should not be confused with an enacted comprehensive incident-response regime.
The distinction matters because autonomous systems complicate responsibility.
If an AI agent causes harm, responsibility may involve the developer, deployer, operator or organization authorizing its use.
The appropriate allocation depends on the facts and applicable law.
Government oversight must therefore address not only the properties of AI models but also the systems and institutions deploying them.
National Security and the Private Model
The national-security implications are particularly significant.
Frontier AI can support cybersecurity, intelligence analysis, scientific research and potentially military-related applications.
Governments increasingly recognize that access to advanced models may provide strategic advantages.
But a capability supplied through a private company can be affected by commercial decisions.
Access conditions may change.
Pricing may change.
Models may be updated or withdrawn.
Service availability may depend on infrastructure located abroad.
Security restrictions may vary between users.
These are ordinary features of commercial technology.
They become strategically important when the technology supports essential government functions.
The challenge is therefore to establish relationships in which public authorities retain sufficient oversight and continuity without unnecessarily restricting innovation.
A New Social Contract for Strategic Technology
The historical relationship between states and strategic industries suggests that a durable solution requires institutional arrangements rather than rhetorical declarations.
Governments need independent technical expertise.
They need access to evidence about system capabilities and risks.
They need clear statutory authorities where intervention is necessary.
They need mechanisms for reviewing the security of AI systems used in public functions.
They need procurement arrangements that preserve accountability.
They need operational alternatives for essential services.
And they need international cooperation because frontier AI development and deployment cross national boundaries.
Technology companies also have legitimate interests.
They need predictable regulation.
They need protection for proprietary information.
They need clear security requirements.
They need incentives to invest in research and safety.
An effective governance framework must reconcile these interests.
The objective should not be to place every technological decision under direct political control.
It should be to ensure that decisions with significant public consequences remain subject to legitimate public authority.
The Limits of National Solutions
No single government can fully resolve the governance challenges of frontier AI.
The technology is developed internationally.
Its supply chains cross borders.
Its models can be deployed through global platforms.
Its capabilities can affect security conditions in multiple jurisdictions.
Uncoordinated national rules may produce regulatory fragmentation.
But complete international uniformity is also unlikely.
States differ in political systems, legal traditions, economic interests and security priorities.
International AI governance will therefore require cooperation among countries that do not share identical preferences.
Possible areas include common evaluation methods, incident-reporting practices, technical standards, scientific cooperation and safeguards against severe misuse.
The objective should be practical interoperability rather than the assumption that every government will adopt the same regulatory model.
The Strategic Choice Facing Britain
Britain’s immediate challenge is to convert technical expertise into durable public authority.
The AI Security Institute provides an important foundation.
Its evaluations can help government officials understand emerging capabilities.
Its partnerships can support international cooperation.
Its research can inform legislation and operational security.
But the long-term credibility of the British approach will depend on whether independent assessment becomes institutionally reliable rather than contingent on voluntary access.
That does not necessarily require copying the European Union’s regulatory model.
Nor does it require rejecting cooperation with American AI developers.
It requires defining the conditions under which the British state can obtain the information and capabilities necessary to fulfill its responsibilities.
The distinction is between cooperation as a useful policy instrument and cooperation as an indispensable substitute for public authority.
Bottom Line Assessment
The central political question surrounding frontier artificial intelligence is changing.
For years, debate focused on whether governments should regulate AI.
Increasingly, the question is whether governments possess the practical capacity to govern systems they do not develop, operate or fully understand.
Britain’s experience illustrates this transformation.
The country has invested in independent AI evaluation.
It is deploying advanced AI for cybersecurity.
It is examining new protections against autonomous-system risks.
And its political institutions are debating the adequacy of voluntary oversight.
Yet its most advanced evaluations still depend substantially on cooperation with private developers and international partners.
This is not evidence that national sovereignty has disappeared.
Nor does it demonstrate that private AI companies have replaced governments.
It reveals a growing difference between possessing legal authority and possessing the technological capacity required to exercise it effectively.
Historically, states addressed similar challenges through regulation, public research institutions, industrial policy and strategic partnerships.
Artificial intelligence will require a comparable institutional response, adapted to a technology that is more general-purpose, internationally distributed and rapidly evolving than many earlier strategic industries.
The future of AI sovereignty will not be determined simply by which country passes the strictest law or develops the most powerful model.
It will depend on whether governments can combine democratic legitimacy, technical understanding, enforceable oversight and operational resilience.
Private innovation and public authority need not be opposing forces.
But their relationship must be structured so that responsibility remains connected to effective control.
The defining sovereignty question of the AI era is not whether governments can command technology companies. It is whether democratic institutions can remain genuinely accountable for technologies on which their societies—and increasingly their own security capabilities—depend.
Vladimir Tsakanyan, Ph.D.
Center for Cyber Diplomacy and International Security (CCD-IS)
Leave a Reply