Anthropic discusses the China-U.S. AI race: China’s lead could become a global threat; three recommendations to strengthen America’s moat

Well-known AI company Anthropic said in its latest published research report that ensuring the United States and its democratic allies maintain leadership in the AI field is key to preventing authoritarian regimes from using advanced technology for large-scale surveillance and military expansion. Anthropic urged the U.S. government to immediately patch loopholes in existing chip export controls and strongly counter China’s theft of R&D achievements through “model distillation” technology, so as to secure a key technological lead of 12 to 24 months by 2028.

AI hashrate has become the core of great-power games; Anthropic calls for blocking technological expansion by authoritarian regimes

In the report, Anthropic emphasized that the most important core resource for developing cutting-edge artificial intelligence is “hashrate,” advanced semiconductor chips used to train models. At present, the democratic camp has an absolute advantage in chip innovation and manufacturing, thanks to the technological strength of the United States and its allies, as well as export-control policies supported by both parties in recent years.

However, China is actively developing the AI field and has already applied the technology to speech censorship, cyberattacks on the internet, and pushing the People’s Liberation Army’s “intelligent” military upgrade. If development standards for AI technology in the future are led by authoritarian regimes, it will pose an unprecedented threat to global security and democratic freedoms.

Uncovering the inside story of China’s AI catch-up: smuggling high-end chips and “model distillation” data-theft attacks

Despite the United States implementing strict export controls, China’s AI labs are still chasing closely behind through various channels. The report reveals that besides having top R&D talent, China more frequently exploits policy loopholes to obtain restricted U.S. high-end chips—for example, through smuggling, or by directly renting cloud compute from overseas data centers such as those in Southeast Asia.

In addition, Chinese companies have also launched large-scale “model distillation attacks,” by setting up large numbers of fake accounts to bypass restrictions and systematically harvest output data from advanced U.S. AI models—so as to “free-ride” by copying the innovation results that the U.S. invested tens of billions of Dollars in and poured vast amounts of R&D effort into at extremely low cost.

A major technical breakthrough in 2026: the Mythos Preview model shows disruptive cybersecurity potential

The pace of AI technology advancement is growing exponentially. Anthropic specifically mentioned the “Mythos Preview” model released in April this year, which can, in just one month, help the Firefox browser repair the number of software security vulnerabilities that exceeds the total for all of 2025. The explosive growth in technology suggests that “super geniuses in the data center” is about to become real, while also highlighting the urgency of policy intervention. If no action is taken immediately, advanced AI being used to automatically search for system vulnerabilities or coordinate large swarms of drones in military threats will increase significantly.

Previewing two future scenarios for 2028: democratic camp-led global norms or authoritarian regime rise

The report presents two sharply different scenarios for the global AI landscape in 2028. In the first scenario, the United States and its allies successfully expand their hashrate advantage, fully block China’s theft activities, and widen the technological lead to 12 to 24 months. Then, U.S. AI technology becomes the backbone of the global economy, with the democratic camp setting AI rules to ensure safety and human rights.

In the second scenario, U.S. policy stalls. China will reach technical parity through policy loopholes and continued theft, and use companies such as Huawei and Alibaba to export low-cost AI infrastructure lacking security protections to markets in the Global South, thereby overturning the existing balance of international power.

Anthropic proposes three major policy solutions, urging a comprehensive strengthening of the U.S. AI moat

To ensure the democratic camp can smoothly move toward the optimistic first scenario, Anthropic put forward three specific policy recommendations at the end of the report.

First is “comprehensively patching loopholes.” It calls on the government to increase law-enforcement budgets, strictly crack down on chip smuggling, block overseas data-center pathways for renting compute power, and strengthen export restrictions on semiconductor manufacturing equipment (SME). Second is “defending innovation achievements.” It suggests that through legislation, “model distillation attacks” be clearly defined as illegal acts, and that technology and intelligence sharing between companies and the government be promoted. Finally is “promoting the export of U.S. AI technology.” It actively promotes AI infrastructure that aligns with democratic values worldwide to seize international market share and prevent the technological ecosystem of authoritarian regimes from expanding globally.

This article about the China-U.S. AI race discussed by Anthropic: the fear that China’s lead could become a global threat, and three recommendations to strengthen the U.S. moat, first appeared on Lian News ABMedia.