AI Safety in the Shadow of US–China Rivalry : Turning a Tech Debate Into a New Cold War

Artificial intelligence (AI) has become difficult to separate from everyday life for much of the world. Its capabilities have advanced rapidly, while its uses have spread across age groups and professions, from children generating images or completing school assignments to employees and scientists using AI as a professional assistant. The technology is no longer confined to specialised laboratories; it is increasingly embedded in ordinary economic and social activity.

Yet rapid adoption also brings risks. In a recent essay, ‘We Must Pace the Frontier’, Dario Amodei, CEO of Anthropic, called for the AI industry to slow the pace of frontier-model development. He argues that AI is a powerful tool that could create serious risks, including loss of control over AI systems, their misuse for cyberattacks and bioterrorism, and severe economic disruption. His concern is partly driven by the speed at which AI capabilities are advancing. Amodei warns that within 6-12 months, increasingly capable AI botnets could potentially gain control over significant parts of the internet, potentially causing hundreds of billions of dollars in damage. He further argues that the scale of disruption could increase as AI systems become more capable without adequate safeguards. The warning has helped bring what was previously a relatively specialised debate over frontier-model safety into a broader public and geopolitical discussion.

That discussion has since moved into government circles. As major powers increasingly view AI as a strategic asset, questions about its development are becoming intertwined with national security and geopolitical competition. The issue is therefore no longer simply how to make AI safer, but how to regulate a technology that governments also regard as a source of economic and strategic advantage.

At the centre of the debate is a tension between safety and technological sovereignty. AI developers warn that increasingly capable systems could generate serious risks, including cyberattacks and disruption to critical infrastructure. US policymakers, however, face another calculation. Slowing frontier AI development could, in their view, risk weakening America's technological lead over China. This creates a difficult incentive structure. Measures intended to reduce technological risks could also be seen as limiting a country's strategic position if its main competitor continues to advance.

Beijing has challenged the way Washington and parts of the Western AI industry frame this competition. Chinese officials and state media have questioned whether calls to slow AI development are motivated solely by safety concerns. The disagreement consequently extends beyond the question of how quickly AI should develop. It also raises the question of who has the capacity and the incentive to determine how far that development should go.

The main actors are therefore no longer limited to technology companies. Amodei is not calling for an outright halt to AI development, but for its pace to be controlled so that safety measures, testing and governance can keep up with increasingly capable systems. In the United States, President Donald Trump has rejected measures that could slow AI development and has emphasised the importance of maintaining America's technological advantage over China. China, meanwhile, has continued to strengthen its domestic AI capabilities while seeking to reduce its dependence on American technology.

Chinese officials and state media have also resisted framing the issue as a straightforward competition over AI safety. Global Times, a state-backed Chinese newspaper, described Amodei's proposal as a “Cold War playbook” aimed at containing China. The response illustrates the central difficulty in the debate, that is an attempt to manage technological risks can be interpreted by a geopolitical rival as an attempt to constrain its strategic development.

The issue has consequently evolved from a question of corporate decision-making into one of national policy and international competition. The underlying disagreement is on whether countries at the technological frontier can coordinate limits on its development when each has an incentive to ensure that its rivals do not move ahead.

For the US and China, that may make cooperation on slowing AI development particularly difficult. Chang Jun Yan, an assistant professor in the Military Studies Programme at Singapore's S. Rajaratnam School of International Studies, argues that cooperation to make AI development slower and safer is “next to impossible” because each country has its own security and political interests. If that assessment is correct, the challenge may not be a lack of awareness of AI's risks, but the strategic incentives that make restraint difficult. ◆