When AI Becomes a Security Weapon: Inside the U.S.–China Race to Control Artificial Intelligence

Kathmandu, Nepal: Artificial intelligence is developing rapidly, and governments are struggling with a difficult question: how can they control a technology they are also competing to develop and use?

The issue is particularly important in the growing rivalry between the United States and China. Both countries see artificial intelligence as an important source of economic and technological power. At the same time, both are increasingly concerned about the ways advanced AI could be used against governments, businesses and the public.

The debate is no longer limited to chatbots or automated software. Generative AI can now produce realistic text, images, audio and video, while newer systems can perform complicated tasks with less human involvement.

These advances could improve research, business, education and public services. But they also create new concerns involving cybersecurity, misinformation, military technology, political influence and the protection of sensitive information.

The United States and China broadly agree that these risks deserve attention. Their disagreement is over how the risks should be handled and who should set the rules.

China warns about AI misuse

China's State Security Ministry has increasingly focused on the security risks associated with artificial intelligence.

Chinese officials have warned that generative AI, deepfakes, fabricated information and automated online campaigns could be used to influence public opinion and create social or political instability.

State Security Minister Chen Yixin has warned that hostile groups and individuals could misuse generative AI. The concerns include manipulated information, cybersecurity threats, the exposure of sensitive data and possible risks to military systems.

Some of these technologies are already widely available.

A deepfake video, for example, can make it appear that a public official said something that was never said. AI-generated images can portray events that never occurred, while automated systems can produce large numbers of messages and posts for distribution across social-media platforms.

The concern grows when these tools are used together.

A campaign can produce different messages for different groups, create accompanying images and videos, translate the material into several languages and distribute it through automated accounts.

Chinese authorities have described this type of activity as “cognitive warfare,” referring to attempts to influence how people understand events and form opinions.

This has broadened the security discussion around AI. The technology is no longer viewed only as a way to automate tasks or produce content. It can also influence the information people receive and the environment in which political and social decisions are made.

The problem of scale

One of the biggest concerns is not necessarily a single fake video or misleading image. It is the ability to produce enormous amounts of fabricated material quickly and cheaply.

Traditional influence operations often require people to write material, create images, manage accounts and distribute content. Generative AI can automate many of those jobs.

A person or organization with malicious intentions could potentially produce different versions of a message for different audiences, generate supporting material, translate it and distribute it through automated networks.

That ability changes the scale of misinformation.

For this reason, governments increasingly regard AI-generated misinformation as a national-security concern rather than a problem limited to social-media companies.

Concerns over sensitive information

AI is also raising questions about data security.

Government agencies, companies and other organizations are using AI tools to analyze documents, conduct research, write computer code and assist with decisions. These systems can be useful because they can process large amounts of information quickly.

But using an outside AI service can create security concerns when confidential information is involved.

An organization that sends sensitive documents or data to an external AI system must rely on that provider's security measures and data-handling practices.

Chinese officials have warned that foreign AI systems could provide a channel through which government, business or other sensitive information could leave protected environments.

The issue is relevant beyond China. Organizations in many countries are facing the same question: how much confidential information should they provide to an AI system that they do not fully control?

For governments, the consequences of a serious data breach could be particularly significant.

The U.S. debate over slowing AI development

In the United States, some technology leaders and researchers are also warning that AI capabilities may be advancing faster than safety measures.

Anthropic CEO Dario Amodei has argued that additional time may be needed for safety research, testing and international discussions before the development of more powerful systems moves further ahead.

The argument is not necessarily that AI development should stop permanently. Rather, supporters of greater caution say governments and researchers need enough time to understand the risks and develop effective safeguards.

According to the supplied material, the proposal has received support from prominent technology figures including OpenAI CEO Sam Altman, Elon Musk and Google DeepMind chairman Demis Hassabis.

But slowing development creates another concern for U.S. policymakers: China may not slow down.

The competition with China

For Washington, AI safety is closely connected to technological competition with Beijing.

Amodei has supported stronger restrictions on China's access to advanced AI chips and chipmaking equipment. He has also raised concerns about efforts to reproduce or improve AI capabilities developed by U.S. companies.

One issue is model distillation, a process through which information or behavior produced by a more capable model can be used to help train or improve another model.

This creates a difficult policy choice for the United States.

If American companies reduce the pace of development while Chinese companies continue moving quickly, Washington could fear losing an important technological advantage.

On the other hand, continuing to develop increasingly capable systems without adequate testing could increase the risks associated with those systems.

The basic conflict is straightforward: safety measures can require additional time, while international competition encourages companies and governments to move quickly.

Beijing questions the U.S. approach

China has also questioned whether calls for greater AI safety are completely separate from the broader U.S.–China technology dispute.

Beijing is particularly skeptical because discussions about AI safety have taken place alongside U.S. restrictions on advanced semiconductors and computing technology available to Chinese companies.

The state-backed Global Times has criticized proposals to slow AI development, arguing that such measures could be used to limit China's technological progress. Its editorial positions, however, should not automatically be treated as official Chinese government policy.

China's Foreign Ministry has separately called for AI development to remain open and inclusive and to serve human interests.

Foreign Ministry spokesperson Guo Jiakun has warned against using concerns about AI risks to create fear, increase confrontation or promote what China considers unfair competition.

The disagreement, therefore, is not simply about whether AI poses risks. Both countries acknowledge that it does.

The more difficult question is whether discussions about AI safety can be separated from the competition over advanced technology.

China is developing its own AI safety measures

At the same time that China criticizes some Western approaches to AI governance, it is developing its own security framework.

According to the supplied material, China has introduced an Artificial Intelligence Security Governance Framework 3.0 and has expanded its planning to cover increasingly capable and potentially autonomous AI systems.

Earlier discussions considered scenarios in which future AI systems could obtain resources independently, reproduce themselves or seek greater control. Later work also examined the possibility of rapid increases in AI capabilities and situations in which advanced systems could become difficult for people to control.

A central principle in these discussions is that humans should retain control over AI systems.

China has also proposed requirements for developers of AI agents to identify harmful or inappropriate behavior, stop dangerous actions and restore human control when necessary.

Those concerns are also being discussed by AI safety researchers in other countries. As AI systems become more capable of acting on their own, reliable ways for people to monitor, interrupt and control them become increasingly important.

Different priorities in Washington and Beijing

The United States and China increasingly recognize many of the same AI-related risks, but their priorities are not identical.

In Washington, much of the discussion focuses on the possibility that advanced AI could become highly capable in areas such as cyber operations, scientific research or other sensitive fields. There are also concerns about AI being used to assist dangerous biological activities.

In Beijing, greater emphasis has been placed on political stability, information security, misinformation, critical infrastructure, sensitive data and dependence on foreign technology.

The difference can be summarized simply.

U.S. policymakers and researchers often focus on what increasingly advanced AI systems might eventually be able to do.

Chinese officials place considerable attention on how AI can already be used by governments, organizations and individuals.

The two concerns can exist at the same time. In fact, as AI develops, they are likely to overlap more often.

A lack of trust makes cooperation difficult

The biggest obstacle to cooperation may be political rather than technical.

The United States is concerned that unrestricted access to advanced computing and AI technology could help China close the technological gap or gain an advantage in areas important to national security.

China, meanwhile, views U.S. restrictions on advanced chips and computing equipment as part of an effort to limit its technological development.

This creates an awkward situation.

Washington wants international discussions about AI safety but is also restricting China's access to some of the technology needed to build advanced AI systems.

Beijing wants a role in international AI governance but remains skeptical of proposals that it believes could reinforce existing technology restrictions.

Asia Society Policy Institute expert Lizzie C. Lee has highlighted this tension, according to the supplied material.

As AI becomes more important to national security and economic policy, resolving that disagreement is likely to become more difficult.

The United States has its own disagreement

The debate inside the United States is also divided.

Some researchers and technology leaders believe AI development should be slowed enough to allow safety research and regulation to catch up.

Others argue that maintaining technological leadership is itself a security priority. From this perspective, slowing U.S. development while competitors continue advancing could create a different and potentially serious risk.

According to the supplied material, President Donald Trump has emphasized maintaining U.S. leadership over China in AI rather than supporting proposals to slow development.

The disagreement reflects the wider difficulty of setting AI policy.

The technology has enormous economic potential, and companies have strong incentives to develop and deploy new systems. At the same time, governments increasingly view AI capability as an element of national power.

Could the world end up with competing AI systems?

If the United States and China cannot find areas for cooperation, the effects could reach other countries.

The world could gradually develop separate AI ecosystems, with different technical standards, security requirements and regulatory systems.

One group of countries could place greater emphasis on open research and commercial innovation, while another could give greater priority to government oversight, national security and information control.

Other governments would then have to decide which systems and standards to adopt.

Such a division could affect cybersecurity, data protection, international trade and scientific cooperation. It could also encourage countries to move faster simply because they fear falling behind their competitors.

That would make it harder to ensure that new AI systems are adequately tested before they are widely deployed.

What could come next?

There is still room for U.S.–China discussions on AI safety.

Possible areas of cooperation include safety testing, reporting serious AI incidents, limiting particularly dangerous military uses, preventing biological misuse and establishing standards for autonomous AI systems.

However, according to the supplied material, the two countries have not yet reached a concrete agreement covering these areas.

That leaves an important problem unresolved.

The countries developing some of the world's most advanced AI systems have strong reasons to protect information about their capabilities. Yet effective safety standards may require a certain degree of transparency and communication.

Finding a balance between national security and international cooperation will be one of the major challenges facing policymakers.

AI competition is becoming a governance issue

For years, the AI competition was largely measured by computing power, model performance, research breakthroughs and commercial adoption.

That is beginning to change.

Governments and companies will increasingly be judged on whether they can operate powerful AI systems securely, protect sensitive information, prevent harmful autonomous actions and identify misuse.

They will also need ways to test advanced systems before deployment and respond when those systems behave unexpectedly.

This means AI governance is becoming part of technological leadership.

Having the most capable model may not be enough. Companies and governments will also need to show that their systems can be used responsibly and that meaningful safeguards are in place.

A competition unlike a traditional arms race

The AI competition does not look exactly like a conventional arms race.

There is no single weapon at its center and no clearly defined point at which the competition ends.

Instead, countries are competing across several areas, including semiconductor technology, data centers, algorithms, research talent, cloud computing and increasingly capable AI models.

Still, some of the underlying dynamics are familiar.

Countries worry about falling behind. They want an advantage over competitors and are concerned about what other governments may develop.

As a result, AI capability is increasingly being treated as part of national security.

The difficulty is that the same technology can have very different uses.

AI can help scientists analyze research, businesses improve productivity and governments deliver public services. It can also be used for cyberattacks, large-scale manipulation of information or military purposes.

That makes decisions about AI development more complicated than a simple competition between technological leaders.

The larger question

Artificial intelligence is advancing faster than governments can establish comprehensive rules for it.

The United States and China both recognize that increasingly capable AI can create serious security problems. Their disagreement is largely about how those problems should be addressed and how much technological competition should be limited in the process.

Technical safeguards alone are unlikely to settle the issue.

Effective AI security will also depend on transparency, reliable testing, international communication and a degree of trust between governments.

The United States and China do not have to agree on every aspect of AI policy to make progress. They could still establish limited agreements covering areas where both sides face similar risks.

Those areas could include testing advanced systems, responding to major AI incidents, protecting sensitive information and maintaining human control over autonomous systems.

The alternative is greater fragmentation, with countries developing separate standards while competing to build increasingly capable technology.

The central issue, therefore, is no longer simply who can build the most powerful AI system.

It is whether governments and technology companies can develop systems that are powerful, useful and secure—and whether competing countries can find enough common ground to prevent AI safety from becoming another source of conflict.