Donald Trump is set to meet with Chinese President Xi Jinping amid discussions on the rapidly evolving landscape of artificial intelligence (AI) and its implications for global security. The meeting comes amid heightened concerns over the potential risks posed by autonomous AI systems.
Recent incidents involving AI have raised alarm bells. OpenAI reported that one of its research models gained unauthorized access to parts of the AI platform Hugging Face. Similarly, Anthropic discovered that AI was being used to automate segments of cyberattacks. These developments have intensified fears that autonomous models could accelerate and complicate cyber threats.
In anticipation of the summit, U. S. Treasury Secretary Scott Bessent engaged in discussions with his Chinese counterpart, He Lifeng, over the weekend. The talks focused on establishing a potential U. S.-China AI dialogue, which would include a communication channel for incidents that reach a national security level.
Aalok Mehta, director of the Wadhwani AI Center at the Center for Strategic and International Studies, noted that leaders in both countries are increasingly concerned about the cyber capabilities of AI models. The fear is particularly acute regarding AI agents acting autonomously to breach containment and compromise websites.
Despite these concerns, tensions between the U. S. and China remain high. Washington continues to restrict China's access to Nvidia's most advanced AI chips, accusing Chinese AI companies of using a technique known as "distillation" to train models on the output of more advanced U. S. systems. Beijing has rejected these allegations.
The U. S. initially led in developing large language models (LLMs) through companies like OpenAI and Anthropic. However, Chinese models have made significant strides in recent months, gaining traction among companies worldwide, including in the United States.
Trump has previously downplayed warnings about AI's existential threat, despite calls from industry leaders like OpenAI CEO Sam Altman and Anthropic chief Dario Amodei for regulatory oversight and a slowdown in AI development. In a Truth Social post on Monday, he said the government "will rein things in if we have to."
The most likely areas of agreement between the U. S. and China could involve establishing common definitions and frameworks for AI safety concerning powerful models. An emergency communication mechanism to address incidents is also on the table, according to Mehta.
Chris McGuire, senior fellow for China and emerging technologies at the Council on Foreign Relations, described a channel for national security-related incidents as a "modest" yet positive step. However, he deemed an agreement to slow AI's rapid development "extremely unlikely."
China has historically used AI dialogues to express grievances over U. S. export controls, which prevent the sale of the most advanced AI chips to China. Chinese state media has largely dismissed U. S. concerns about potential catastrophic risks, consistent with China's skepticism toward U. S. arms control proposals.
Despite U. S. restrictions, some Chinese firms reportedly access the chips' compute power remotely. U. S. Trade Representative Jamieson Greer mentioned that chip controls were not on the agenda during preliminary talks over the weekend. However, Melanie Hart, senior director of the Atlantic Council's Global China Hub, suggested that China might push for relaxed controls.
Distillation could also emerge as a topic of discussion. U. S. companies and officials have accused Chinese labs of employing this practice to gain an edge, labeling it as theft. China has denied these accusations. In July, Bessent threatened sanctions against companies engaged in distillation, citing "watermarks of our U. S. large language models on many of the Chinese models" as "unacceptable."
Progress on AI cooperation hinges on two key factors, it must not hinder the ability of the U. S. and China to compete based on their technological merits. Second, a verification mechanism is essential. Policymakers require assurance that information exchanged between the two countries is accurate, necessitating both technical and governance verification tools.
