Concerns Over AI Models and Potential Sanctions

Treasury Secretary Scott Bessent expressed concerns on Tuesday about the possibility of imposing sanctions on a Chinese AI company if evidence emerges showing that its open-source AI models were developed using stolen data from U.S. laboratories.

“We’re in favor of an open source model, but not when it comes to intellectual property theft,” he stated in his address. “If we find that foreign companies are taking from our esteemed firms, we can take action against them.”

Recent reports indicate that major AI firms like Anthropic, OpenAI, and Google have raised alarms about labs in China using unauthorized distillation methods. These methods supposedly allow newer models to be trained without permission, utilizing more advanced existing models.

As powerful open-source models like Moonshot’s Kim K3 gain traction in the U.S., concerns are growing, particularly as they come at a much lower price than those from companies like Anthropic and OpenAI.

Bessent noted that U.S. authorities have discovered “watermarks” of American AI technology embedded in models produced in China, describing this trend as “unacceptable.”

“We’ll be investigating this further in the coming days and weeks,” he added.

Anthropic has accused Moonshot of employing distillation techniques as early as February, though it’s yet to confirm whether there’s solid evidence connecting Kimi K3 to this practice.

Looking ahead, Bessent is set to lead discussions between U.S. and Chinese officials focused on artificial intelligence in September, mentioning that U.S. companies might face new disclosure requirements if they utilize Chinese models.

His remarks come amid ongoing conversations among senior U.S. officials regarding the best strategies to respond to China’s swift progress in AI technologies.

The belief that U.S. laboratories maintain a 6-12 month lead over their Chinese counterparts is increasingly questioned, especially following the success of Moonshot and another company, Z.ai, which launched an open-source model called GLM-5.2 earlier this month.

Additionally, Axios reported earlier this week that the Trump administration is weighing options to restrict the Chinese model, including possibly adding Chinese research facilities to the Commerce Department’s Entity List. This would mean that, without a license, they could not conduct business in the U.S.

OpenAI’s director of strategic futures, Dean Ball, stirred controversy by stating on social media that the rise of open-source AI models might lead to what he termed “full-fledged AI communism,” suggesting a parallel to China’s proposed approach.

Critics argue that such regulatory claims could stifle fair competition in the AI sector.

White House AI adviser David Sachs countered Ball’s assertions in a detailed post, implying that Ball might be attempting to shape regulations to benefit OpenAI.

Sachs also pointed out that the U.S. is “at a critical inflection point in AI policy,” emphasizing that larger, closed labs already dominate revenue from AI models and may seek to eliminate open-source competition.