US, China and artificial intelligence: The Economist on a cold war-style dilemma

When Xi Jinping and Donald Trump meet in Beijing on May 14–15, artificial intelligence may be added to the traditional set of complex topics — the war in the Middle East, trade imbalances and the status of Taiwan.

The Economist writes about this in an article published under the headline “AI creates a fearsome cold-war-style dilemma.”

According to the publication, elites in Beijing and Washington are increasingly alarmed by the speed of technological development: the smarter the models become, the more important they are for the economy and the country’s geopolitical weight — but at the same time the risks associated with them are also growing.

“Since the creation of the atomic bomb, great powers have not faced such a dilemma,” The Economist notes.

The reason for concern was the Mythos model. The publication reports that in April the American laboratory Anthropic (creators of the AI Claude) announced the creation of a model so effective at finding vulnerabilities in cybersecurity that it was decided not to publish it. The reaction was international: after initial skepticism, Chinese state media noted Mythos’s “unprecedented capabilities for cyberattacks,” and one Russian television channel called it “worse than a nuclear bomb.”

According to The Economist, after this episode the Trump administration abandoned its previous approach of non-interference in technology regulation and is considering the possibility of mandatory government review of new models.

The strategic positions of the parties differ

As The Economist writes, part of the American technology community believes that the country that first creates a self-improving AI model will gain an enormous strategic advantage. Chinese experts, however, tend to view AI as the key to economic growth — “more like nuclear energy than nuclear weapons,” the publication formulates. At the same time, neither side is ready to slow its own development, fearing to cede an advantage to its rival.

Voices in favor of cooperation are heard from both sides

At an event in the Capitol at the end of April, Chinese government adviser Xue Lan stated that the United States and China have a “mutual interest” in AI safety. “If one country is not safe, then none of us are safe,” The Economist quotes him as saying. Xue Lan, together with Zeng Yi, head of the Beijing Institute for AI Safety, called for global efforts to regulate and even slow the development of AI. This, the publication notes, is consistent with Beijing’s official line: after the appearance of ChatGPT in 2022, China has been promoting the idea of international cooperation and the creation of a body under the UN, and Chinese diplomats have floated the idea of a “pause” in AI development.

Three possible formats of cooperation

The first is dialogue, or “strategic reassurance” analogous to nuclear arms negotiations: countries could write rules in parallel, without formal coordination.

“If they read the same technical papers and have a similar view of reality, both sides can take reasonable countermeasures,” the publication quotes Carson Elmgren of the Institute for AI Policy and Strategy.

The second format is the harmonization of common methods for testing the safety of models without exchanging results.

The third, most ambitious, is a formal agreement on joint tests and the exchange of their results with inspections modeled on the IAEA.

Obstacles to cooperation still outweigh incentives

American researchers, according to The Economist, are skeptical about the sincerity of Chinese officials: Chinese laboratories are relatively weak in safety work, and the documentation for the recently released DeepSeek v4 model did not include the protective measures adopted in American laboratories. Former Biden administration official Ryan Fedasiuk, according to the publication, believes that China’s statements on AI safety are demonstrative and are made, in his words, “to put Americans in an unfavorable light.”

The Economist also recalls that in 2024 in Geneva the United States sent senior security officials and technical specialists to the negotiations, while China sent political representatives who refused to discuss AI safety until American export restrictions on advanced chips were lifted. Carnegie expert Tong Zhao explains Beijing’s caution by historical memory: the experience of the “unequal treaties” of the 19th century makes Chinese officials suspicious of agreements with a technologically stronger partner.

The stakes of the parties are marked in extremely tough terms

The US Treasury secretary stated in April to the Wall Street Journal: “If we do not win in AI, it is game over.”

Xi Jinping called AI “epoch-making.”

However, as The Economist concludes, leaders of both countries may soon begin to perceive AI as existential for other reasons.

“Unfortunately, historical experience shows that real momentum appears only after a tragic accident,” the publication quotes Jeffrey Ding of George Washington University, recalling the global standards adopted after the Bhopal disaster in 1984 and the Chernobyl nuclear accident in 1986.

“Negotiations on AI remain a real test of human intelligence,” The Economist summarizes.