The competition between the United States and China over artificial intelligence is no longer confined to commercial innovation or industrial policy. It is increasingly being framed in Washington as a national security contest, with cyber capability at its centre. In a recent Financial Times article, Chris McGuire, a senior fellow for China and emerging technologies at the Council on Foreign Relations and a former director for technology and national security at the US National Security Council, argued that America has only a limited period in which to prepare for China’s next advance.

The focus of McGuire’s warning is Anthropic’s Claude Mythos Preview, an advanced AI model whose cyber capabilities have prompted concern among banks, regulators and national security officials. According to the Financial Times, Claude Mythos is capable of autonomously discovering, linking and exploiting software vulnerabilities, as well as helping to patch them. That makes it significant not merely as a coding assistant, but as a tool with potential consequences for cyber defence and cyber operations.

McGuire’s argument is that China is unlikely to remain behind for long. In a public summary of his article, he wrote that Chinese models are only around seven months behind leading US systems. That estimate, if accurate, would leave Washington with a narrow window in which to strengthen banks, hospitals, energy systems, government networks and other critical infrastructure before comparable Chinese tools become available.

The risks are already visible. In November 2025, Anthropic said it had disrupted what it described as the first reported AI-orchestrated cyber espionage campaign. The company said the operation used AI not only as an advisory tool, but to carry out parts of the cyber operation itself. It attributed the campaign with high confidence to a Chinese state-sponsored group and said the targets included technology companies, financial institutions, chemical companies and government agencies.

Reporting by The Associated Press said the campaign targeted about 30 entities worldwide and succeeded in a small number of cases. Anthropic said the attackers manipulated Claude by presenting themselves as legitimate cybersecurity professionals. The case demonstrated one of the central problems of advanced AI: systems designed to help defenders identify vulnerabilities can also be used by hostile actors to accelerate attacks.

The strategic concern for Washington is that China’s AI progress remains deeply linked to American and allied technology. Chinese companies continue to seek access to advanced AI chips, cloud computing capacity, US-developed models and semiconductor manufacturing equipment. McGuire argues that without these inputs, Chinese AI development would be delayed by years rather than months.

The semiconductor dimension is particularly important. The United States has already imposed controls on advanced chip exports to China, while the Netherlands and Japan have restricted exports of some semiconductor manufacturing equipment. Yet enforcement remains difficult. Smuggling networks, third-country purchases and access through cloud services can all reduce the practical effect of controls. McGuire therefore proposes a far stricter regime.

Anthropic’s White House return exposes Trump’s muddled AI policy

His recommendations include blocking all AI chip exports to China, including less advanced models; requiring export licences for large chip orders globally; closing routes through cloud platforms; restricting Chinese access to advanced US AI models through both export and remote access; and preventing the export of equipment capable of supporting advanced semiconductor production where US components are involved.

The aim, in his view, is not simply to slow China. It is to widen the American lead from months to 18 months or more. That additional time would allow the US to deploy AI-enabled defensive tools across public and private digital infrastructure before China develops equivalent offensive capabilities. Every additional month, in this argument, could mean another bank, hospital, power grid or government system better protected against AI-assisted intrusion.

McGuire also draws a comparison with the Cold War. The United States sought to prevent the Soviet Union from acquiring sensitive nuclear technology, while later engaging on safeguards designed to reduce the risk of unauthorised launches. He suggests a similar approach for AI: preserve technological advantage while also negotiating rules and limits with Beijing.

For Europe, this debate is not distant. European financial institutions, hospitals, ports, telecoms networks and energy systems are part of the same digital environment. A breakthrough in AI-enabled cyber operations would not respect national borders. If advanced models lower the level of skill needed to conduct sophisticated attacks, the number of capable actors could increase and the time available for defenders could fall sharply.

The issue also highlights Europe’s dependence on decisions made in Washington, Silicon Valley and the semiconductor supply chain across the United States, Taiwan, Japan and the Netherlands. While the EU has invested heavily in digital regulation, the coming challenge may be less about rule-setting and more about operational resilience.

McGuire’s warning is that the AI cyber race is now being measured in months. Whether the seven-month estimate proves exact or not, the broader direction is clear. Artificial intelligence is becoming embedded in cyber conflict. The states and companies that adapt first will be better placed to withstand the next phase of digital competition.

First published on euglobal.news.