Washington’s restrictions on China’s DeepSeek and Beijing’s diplomatic response reveal a new stage in technological competition. The strategic boundary between the United States and China is moving beyond semiconductors and telecommunications equipment into the artificial intelligence models themselves.
October 20, 2026 | Cyber Diplomacy & International Security
By Vladimir Tsakanyan, Ph.D.
Center for Cyber Diplomacy and International Security (CCD-IS)
A New Boundary in the Digital World
During the Cold War, technological competition between the United States and the Soviet Union was shaped by a system of restrictions designed to prevent strategically important technologies from crossing geopolitical boundaries.
Advanced computers, electronic components, aerospace systems and specialized industrial equipment became instruments of national power.
Governments increasingly treated access to technology as a matter of strategic security rather than ordinary commercial exchange.
Seven decades later, a comparable political logic is emerging around artificial intelligence.
But the object of restriction has changed.
It is no longer only the semiconductor used to train an AI model, the manufacturing equipment used to produce that semiconductor, or the telecommunications infrastructure through which information travels.
Increasingly, it is the AI model itself.
On October 20, 2026, China publicly criticized the United States over restrictions affecting DeepSeek, one of China’s most prominent artificial intelligence developers.
The immediate dispute concerns an American order excluding DeepSeek products from federal civilian information technology systems and procurement.
Its broader significance lies in what the order represents.
Washington is treating the origin, development and potential data exposure of an AI model as matters of national security.
Beijing sees the measure as another example of technological competition being transformed into political discrimination against Chinese companies.
The confrontation raises a question that extends far beyond DeepSeek:
Can the world’s two leading AI powers establish meaningful rules for cooperation while increasingly restricting each other’s technologies from sensitive institutions?
The answer may determine whether international AI governance develops through shared standards or competing technological spheres of influence.
What Actually Happened on October 20
The chronology matters.
The Department of Homeland Security’s DeepSeek exclusion order originated in July 2026. Its publication in the Federal Register on October 20 made the formal measure publicly accessible and prompted renewed international attention.
The order, issued under the Federal Acquisition Supply Chain Security Act of 2018, addresses two categories of activity.
First, it excludes products and services developed by DeepSeek and its covered affiliates from procurement by federal civilian agencies.
Second, it requires the removal of DeepSeek-related products and certain third-party products or services developed using DeepSeek implementations from federal civilian information technology systems.
The government identified supply-chain and information-security concerns as the basis for the restriction.
The measure is significant because it extends beyond employees simply accessing a particular chatbot.
Its wording also reaches covered third-party implementations, reflecting concern about AI components embedded within larger software environments.
Nevertheless, the legal scope must be distinguished from a general prohibition on American access to Chinese AI.
The order does not, by itself, establish a nationwide ban on private citizens using DeepSeek.
Nor does it prohibit every form of AI research involving Chinese-developed models.
It is a federal civilian information technology and procurement restriction, issued through a specific statutory mechanism.
China’s response was sharply critical.
Foreign Ministry spokesperson Mao Ning opposed what Beijing characterizes as the expansion of national-security justifications into technological and commercial affairs.
China’s Commerce Ministry likewise criticized the measure, arguing that it undermines Chinese business interests and contradicts the cooperative spirit of previous bilateral understandings.
The United States and China are therefore disputing not only the security of one AI developer.
They are disputing the legitimacy of using national-security authority to exclude a competitor’s technology.
The Historical Origins of Technological Containment
The political logic behind technological restrictions predates the internet.
In 1949, Western governments established the Coordinating Committee for Multilateral Export Controls, commonly known as COCOM.
Its purpose was to coordinate restrictions on exports of strategically important goods and technologies to the Soviet bloc.
Advanced computing and electronics were among the technologies subjected to controls.
The underlying assumption was straightforward.
Scientific and industrial capabilities could strengthen an adversary’s military and intelligence systems.
Restricting access could therefore slow the development of strategically dangerous capabilities.
COCOM was not a universal prohibition on trade with communist countries.
It was a selective control system intended to limit transfers of technologies considered sensitive.
Its significance was institutional.
It converted technological interdependence into an object of alliance coordination.
Export policy became a component of national-security strategy.
The Cold War system eventually disappeared, and COCOM was dissolved in 1994.
Its successor arrangements, including the Wassenaar Arrangement established in 1996, reflected a different international environment.
But the central question remained unresolved.
How should governments manage technologies that possess both civilian and military applications?
Artificial intelligence brings that historical problem into a new technological environment.
Unlike many Cold War strategic goods, AI capabilities can be distributed through software, model weights, APIs and cloud services.
The strategic asset may be a downloadable file or a remotely accessible computational capability.
The traditional distinction between controlling physical exports and regulating information becomes increasingly difficult to maintain.
From Semiconductor Controls to Model Restrictions
Modern U.S.–China technological competition initially concentrated heavily on physical infrastructure.
Washington adopted increasingly restrictive policies concerning advanced semiconductors, semiconductor manufacturing equipment and related technologies.
The strategic objective was to limit access to capabilities that could support advanced military computing, surveillance, intelligence and artificial intelligence.
The semiconductor supply chain offered identifiable points of control.
Advanced chips required specialized manufacturing processes.
Those processes depended on equipment, software and technical expertise concentrated among a relatively small number of countries and companies.
Export restrictions could therefore target identifiable suppliers and technologies.
AI models present a different challenge.
Once developed, a model may be accessed through a commercial service, deployed on private infrastructure or distributed under an open-weight license.
Its capabilities can spread without requiring every user to obtain the original training hardware.
DeepSeek illustrates the problem.
Its models have attracted international attention partly because their distribution and deployment options differ from those of proprietary AI systems.
An organization may use a hosted service operated by a developer.
It may also, depending on the particular model and license, deploy model weights on infrastructure under its own control.
These are materially different security arrangements.
A hosted AI service can expose prompts, uploaded documents and usage information to the service operator.
A locally deployed model may reduce that particular exposure while retaining other risks, including vulnerabilities in the software stack, model behavior and supply-chain integrity.
The distinction is essential.
A model’s country of origin, its hosting location and its actual data flows are related but not identical security characteristics.
A rigorous assessment must examine each separately.
Why DeepSeek Became a Security Question
The American concern surrounding DeepSeek centers on potential exposure of sensitive information and supply-chain risks.
When government employees use external AI services, they may submit documents, technical details, source code or other information.
Depending on the service architecture and contractual safeguards, such information may be processed or stored outside government-controlled systems.
Foreign jurisdiction can introduce additional legal and intelligence concerns.
These risks are not unique to Chinese AI services.
Governments face similar questions whenever they rely on external technology providers, including domestic companies.
The difference is the geopolitical and legal environment in which those providers operate.
Washington views certain foreign technology dependencies as potentially exposing sensitive systems to adversarial access or influence.
Beijing rejects what it regards as discriminatory treatment of Chinese companies.
Neither position should be confused with a public demonstration that every DeepSeek implementation has caused a security incident.
The existence of a government restriction establishes that an authorized security determination was made.
It does not independently prove that all conceivable deployments of the technology present identical risks.
That distinction matters for both technical accuracy and diplomatic credibility.
The Huawei Precedent
The DeepSeek controversy follows an earlier confrontation over telecommunications infrastructure.
During the late 2010s and early 2020s, the United States intensified restrictions affecting Huawei, arguing that the company’s role in critical communications infrastructure raised unacceptable national-security risks.
American officials encouraged allies to reconsider dependence on Chinese telecommunications equipment.
China disputed the allegations and criticized the restrictions as politically motivated.
The resulting controversy extended beyond bilateral trade.
It affected procurement policies, allied security cooperation, telecommunications standards and diplomatic relations.
The Huawei dispute established a powerful precedent.
A technology company’s national origin and perceived exposure to foreign-state influence could become decisive factors in determining whether its products were permitted within sensitive infrastructure.
DeepSeek represents a possible extension of that logic.
Telecommunications restrictions primarily concerned the networks through which information moves.
AI restrictions increasingly concern the systems that process information, generate outputs and support institutional decisions.
The distinction is strategically important.
Telecommunications infrastructure provides connectivity.
Artificial intelligence can influence how organizations interpret information and act upon it.
This makes questions of trust, auditability and institutional dependence even more politically sensitive.
The Problem of the Algorithmic Supply Chain
Traditional supply-chain security examines the provenance and integrity of hardware, software components, updates and service providers.
AI complicates that model.
A modern AI application may depend on several layers.
Its underlying model may have been developed by one organization.
Its weights may be distributed through another platform.
Its inference infrastructure may be operated by a third provider.
Its user interface may be developed by a separate company.
Its outputs may be connected to databases, software tools or autonomous workflows.
A restriction on a particular AI developer therefore creates difficult questions about downstream implementations.
How should government agencies identify covered components?
What constitutes a product developed using a restricted implementation?
How should organizations assess a model that has been modified or adapted?
Can the security characteristics of a locally deployed model be distinguished from those of the developer’s hosted service?
These questions require technical and legal precision.
Overly narrow restrictions may fail to address relevant dependencies.
Overly broad restrictions may unnecessarily disrupt legitimate research, software development and government operations.
The challenge is not simply identifying a company’s nationality.
It is determining which technical relationships create unacceptable risks.
The American National-Security Argument
From Washington’s perspective, the restriction reflects a broader principle of defensive governance.
Federal agencies have a responsibility to protect sensitive information and maintain secure information technology systems.
Government procurement is not an unrestricted marketplace.
Security requirements can legitimately affect which products public institutions purchase or operate.
The Federal Acquisition Supply Chain Security Act provides a statutory framework for addressing specified supply-chain risks.
Under this approach, excluding a technology from sensitive government systems is a defensive measure rather than a prohibition on international scientific exchange.
The American argument is strongest when it identifies specific risk pathways, explains the scope of affected systems and applies restrictions proportionately.
It becomes more difficult to defend internationally when technical risk assessments are not sufficiently transparent or when similar risks appear to receive different treatment depending on a supplier’s nationality.
National-security decisions may necessarily rely on classified evidence.
But the absence of publicly accessible evidence can also create diplomatic disputes over the legitimacy of restrictions.
That tension is unlikely to disappear.
The Chinese Diplomatic Argument
Beijing approaches the issue through a different framework.
Chinese officials argue that the United States increasingly invokes national security to justify measures that disadvantage Chinese technology companies.
From this perspective, the DeepSeek restriction is part of a larger pattern involving semiconductors, telecommunications, investment controls and advanced computing.
China portrays these measures as barriers to fair competition and obstacles to stable economic relations.
The October 20 response also reflects Beijing’s interest in preserving the possibility of bilateral cooperation.
Chinese officials have emphasized existing understandings between the two governments and the importance of continued dialogue.
Yet China’s own technology policies also involve national-security considerations.
Beijing maintains controls and regulatory requirements affecting sensitive technologies, data and digital services.
Its recent proposals concerning the export of advanced technologies, including certain AI capabilities, demonstrate that China also regards technological access as a matter of strategic policy.
This does not establish equivalence between the two countries’ legal systems or specific measures.
It demonstrates that both governments increasingly view advanced technology through the lens of national power.
The central diplomatic disagreement concerns which restrictions are legitimate, how they should be justified and whether they can coexist with international cooperation.
The Difference Between a Ban and a Technological Embargo
The phrase AI Iron Curtain is analytically useful, but it requires qualification.
The Cold War Iron Curtain described a profound geopolitical and institutional division.
The current AI landscape remains far more interconnected.
American and Chinese researchers continue to contribute to international scientific knowledge.
Open-weight models circulate across borders.
Universities collaborate.
Commercial software incorporates technologies originating in multiple jurisdictions.
International standards bodies continue to operate.
The October DeepSeek order does not end those relationships.
Nor does it establish a comprehensive embargo on Chinese AI.
Instead, it illustrates the emergence of selective technological exclusion.
Certain AI capabilities may remain available to researchers, companies and the public while becoming unacceptable within sensitive government environments.
The emerging boundary is therefore institutional rather than universally territorial.
It separates trusted and restricted uses.
Over time, however, multiple selective restrictions can accumulate.
Government procurement rules may influence private-sector standards.
Security guidance may shape corporate purchasing decisions.
Cloud providers may adapt their offerings.
Allies may adopt similar requirements.
A collection of individual restrictions can gradually produce a broader technological division without any single decision formally creating one.
That is the strategic danger of incremental fragmentation.
Open-Weight AI and the Limits of Exclusion
DeepSeek’s open-weight distribution creates a particularly difficult policy question.
Unlike a purely proprietary cloud service, an open-weight model can potentially be downloaded, examined, modified and operated independently.
That does not eliminate security risks.
But it changes their nature.
A government might reasonably object to sending sensitive data to a foreign-hosted service while reaching a different technical conclusion about a model deployed within a controlled environment.
Local operation does not automatically make a model trustworthy.
Security still depends on model integrity, software dependencies, deployment architecture, access controls and monitoring.
Nevertheless, treating all forms of use as technically identical can obscure meaningful differences.
This is where AI governance must become more sophisticated than ordinary geopolitical labeling.
The relevant question is not only who created the model.
It is also what the model can access, how it operates, what information leaves the environment and whether its behavior can be independently evaluated.
The Diplomatic Cost of Mutual Exclusion
Technological restrictions can create strategic benefits.
They may reduce exposure to foreign intelligence risks.
They may protect sensitive institutions.
They may limit dependence on suppliers regarded as untrustworthy.
But restrictions also generate diplomatic costs.
They can reinforce suspicion.
They may encourage retaliation.
They can weaken scientific cooperation.
They may complicate negotiations on common safety standards.
They can also encourage competing technological ecosystems.
The resulting problem is particularly serious for AI governance.
The United States and China have strong reasons to cooperate on preventing severe AI misuse, managing autonomous-system risks and reducing the possibility of dangerous incidents.
Yet they also have strong incentives to preserve strategic advantages.
A government may simultaneously seek dialogue with a rival and restrict that rival’s technology from sensitive systems.
There is no automatic contradiction in doing so.
Diplomacy has often continued between strategic competitors.
But cooperation becomes harder when each side interprets defensive measures as evidence of hostile intent.
Lessons From Cold War Arms Control
The Cold War provides an important corrective to the assumption that strategic competition makes cooperation impossible.
The United States and Soviet Union maintained extensive technological and military competition.
They also negotiated arms-control agreements and developed crisis-management mechanisms.
The 1963 Partial Test Ban Treaty, the 1972 Anti-Ballistic Missile Treaty and subsequent strategic arms negotiations demonstrated that rival powers could establish limited agreements despite profound political disagreements.
These arrangements did not eliminate rivalry.
They sought to manage particularly dangerous consequences.
AI diplomacy may require a comparable approach.
The objective need not be unrestricted technological exchange.
It could instead involve cooperation on narrowly defined risks.
Examples include communication mechanisms for serious AI incidents, common evaluation terminology, safeguards against dangerous misuse and procedures for clarifying the intent of certain high-risk activities.
The analogy has limits.
AI systems are developed extensively by private organizations, change rapidly and serve numerous civilian purposes.
They are not directly equivalent to nuclear weapons.
Nevertheless, the diplomatic principle remains relevant.
Strategic competition and risk reduction are not mutually exclusive.
A New Challenge for American Allies
The DeepSeek controversy also has implications beyond Washington and Beijing.
American allies must decide how to assess AI technologies developed in countries regarded as strategic competitors.
Some governments may align closely with U.S. restrictions.
Others may prefer independent technical evaluations.
Countries with limited domestic AI industries may face difficult trade-offs between security, cost and access to advanced capabilities.
Reuters’ October 20 review documents a range of national responses to DeepSeek, including government-use restrictions, privacy investigations and other regulatory measures.
These responses are not identical.
Some concern public-sector devices.
Others concern data protection.
Still others involve broader security assessments.
This diversity demonstrates that the international community has not adopted a single regulatory approach.
For middle powers, the challenge is particularly acute.
They may depend on American cloud infrastructure while seeking access to Chinese open-weight models.
They may cooperate with Washington on cybersecurity while maintaining significant commercial relationships with Beijing.
AI procurement decisions can therefore acquire diplomatic consequences.
A choice that appears technical within a government information technology department may become part of a country’s broader foreign-policy alignment.
The Future of Technological Nonalignment
The emerging AI divide could encourage a new form of technological nonalignment.
During the Cold War, many countries sought to preserve political independence from competing power blocs.
In the AI era, a similar strategy may involve diversification across technology suppliers, investment in domestic evaluation capabilities and resistance to excessive dependence on any single ecosystem.
But technological nonalignment is difficult to achieve.
Advanced AI depends on concentrated semiconductor manufacturing, specialized computing infrastructure, large-scale investment and technical expertise.
A country may be politically independent while remaining economically dependent on foreign technology.
Consequently, effective technological autonomy may require cooperation among smaller and middle powers.
Shared computing infrastructure, independent evaluation institutions and interoperable standards could provide alternatives to exclusive dependence on either Washington or Beijing.
Such arrangements would not eliminate geopolitical pressure.
They could, however, expand the range of available policy choices.
Cyber Diplomacy Beyond Traditional Cybersecurity
The DeepSeek dispute demonstrates why cyber diplomacy can no longer be confined to malicious network activity.
Traditional cybersecurity diplomacy often concerns cybercrime, espionage, critical infrastructure attacks, confidence-building measures and norms of responsible state behavior.
AI introduces additional questions.
Which foreign-developed models can be trusted within public institutions?
How should governments evaluate software supply-chain risks involving AI?
When do national-security restrictions become instruments of economic coercion?
How should states distinguish legitimate defensive measures from discriminatory barriers?
Can international standards survive the fragmentation of technology markets?
These questions connect cybersecurity with trade policy, industrial strategy, international law and alliance politics.
They also create new responsibilities for diplomatic institutions.
Technical agencies may identify risks.
Trade authorities may impose restrictions.
Foreign ministries must manage the resulting political consequences.
The effectiveness of national policy depends on coordination among these institutions.
The Need for Risk-Based Diplomacy
An effective response should begin by distinguishing different categories of technological risk.
A foreign-hosted AI service that receives sensitive government information presents one type of concern.
A locally deployed model with no external data transmission presents another.
An AI application connected to critical infrastructure may present more serious operational consequences.
A model used for low-risk public research presents a different security profile.
These distinctions do not necessarily determine whether a restriction is legally required.
But they are essential to understanding the security rationale and consequences of a policy.
Governments should also distinguish among enacted measures, proposed restrictions and political statements.
The October 20 DeepSeek controversy concerns an issued federal order and subsequent diplomatic criticism.
It should not be confused with hypothetical future restrictions on all Chinese AI technology.
Nor should political warnings be treated as proof that retaliation has already occurred.
Accurate distinctions are especially important in periods of strategic tension, when uncertainty can amplify political mistrust.
Toward a Framework for AI Confidence-Building
The international community may eventually need confidence-building mechanisms tailored specifically to AI supply-chain security.
Such mechanisms could include technical communication channels between relevant authorities, common approaches to evaluating model deployments and procedures for exchanging information about serious vulnerabilities.
They might also establish clearer distinctions between restrictions based on demonstrable technical risks and broader economic measures justified by strategic competition.
These possibilities remain policy options rather than existing bilateral commitments.
Their feasibility would depend on political trust, legal compatibility and willingness to exchange sensitive information.
Some security disputes may remain irreconcilable.
But limited cooperation could still reduce misunderstandings.
The history of international security suggests that durable diplomacy often begins not with agreement on political intentions, but with agreement on procedures for managing dangerous uncertainty.
Bottom Line Assessment
The October 20, 2026 dispute over DeepSeek represents more than another episode in U.S.–China technology competition.
It illustrates a structural shift in the meaning of technological security.
During the Cold War, strategic restrictions concentrated heavily on physical technologies and industrial capabilities.
During the early digital era, governments focused increasingly on communications infrastructure, network equipment and data flows.
Now, the boundary of national security is extending into the models that process information and perform cognitive tasks.
The American government has used an existing supply-chain security framework to exclude DeepSeek-related technologies from federal civilian information systems.
China has responded by challenging the legitimacy of the restriction and portraying it as part of a broader pattern of technological discrimination.
Both positions reflect deeper strategic concerns.
Washington seeks to protect sensitive institutions from perceived foreign technology risks.
Beijing seeks to prevent its emerging AI industry from being excluded from international markets and institutions.
Neither objective can be understood independently of the broader geopolitical competition between the two countries.
The most consequential development, however, may be institutional.
AI models are becoming objects of diplomatic trust and distrust.
Government procurement decisions increasingly carry foreign-policy significance.
Technology standards are becoming instruments of strategic alignment.
And international AI cooperation is being negotiated in an environment where access itself is becoming contested.
The result is not yet a comprehensive technological Iron Curtain.
Global AI research, software distribution and commercial exchange remain deeply interconnected.
But the direction of policy is increasingly clear.
The emerging AI divide may not be built through a single prohibition or treaty. It may emerge through hundreds of procurement decisions, security assessments, export controls and diplomatic disputes that gradually determine which technologies are trusted, where they can operate and whose institutions may depend on them.
The central challenge for cyber diplomacy is therefore not simply to prevent technological separation.
It is to ensure that legitimate security concerns do not eliminate the channels of communication and cooperation needed to manage the risks of increasingly powerful artificial intelligence.
The future of international AI security will depend not only on which country develops the most capable algorithms, but on whether strategic competitors can establish rules for a world in which they no longer fully trust each other’s technology.
Vladimir Tsakanyan, Ph.D.
Center for Cyber Diplomacy and International Security (CCD-IS)
CyberCenter.space
Leave a Reply