Kimi: Friend or Foe? – Tech Weekly

This week, Moonshot AI, a Chinese company, introduced its latest Kimi model, rekindling conversations about the impact of open-source AI in China.

Moonshot noted that Kimi K3 “lags behind the leading proprietary models, Claude Fable 5 and GPT 5.6 Sol,” while claiming that the new open-source model “showed outstanding performance in our evaluation suite, consistently outperforming other models we tested.” Evaluations from Arena.ai and Vals AI suggested that Kimi holds its ground against top frontier models.

The announcement coincided with a speech from Chinese President Xi Jinping at the World AI Conference in Shanghai, which seemingly unsettled Wall Street; the Nasdaq fell by about 1% on Friday as investors moved away from tech stocks, particularly Nvidia.

Reactions from tech leaders echoed earlier sentiments following the January 2025 launch of the open-source R1 model by another Chinese company, DeepSeek. Yet, the current environment appears more fraught due to tariff disputes from the Trump administration, ongoing national security discussions around Anthropic, and the expected public offerings of significant AI companies.

David Sacks, former AI chief during the Trump administration and now co-chair of the President’s Council of Advisors on Science and Technology, praised Kimi’s advancements by juxtaposing it against the United States, which he described as “tying itself in knots: politicians and bureaucrats are banning new data centers, enforcing state regulations, and proposing new federal agencies for pre-approving frontier models. This is how you lose the AI race.” (His remarks also provided a platform to critique Anthropic, branding Claude as a “woke lobotomized model.”)

Former Uber CEO Travis Kalanick expressed worries that the Chinese are “distilling off” (i.e., training on the outputs of) American AI models.

“If distillation isn’t restricted, then everyone should be allowed to distill from anyone else… otherwise, one arm [would be] tied behind American models’ backs,” Kalanick stated. (It’s important to remember that American models have also drawn inspiration from Chinese predecessors, notably Kimi.)

Meanwhile, Dean Ball, head of strategic futures at OpenAI, remarked that Kimi is “a very good model,” and its performance “probably can’t be solely attributed to distillation or similar techniques,” expressing surprise that the Chinese government still allows the open-sourcing of such models given the associated risks.

Ball cautioned that “the likely result of a world dominated by open-weight models is a form of AI communism,” presenting AI as “a ‘public good’ ultimately provided by the state as a type of ‘digital public infrastructure.’”

“This future seems like a dystopian nightmare, yet I haven’t encountered an advocate of open-weight models who doesn’t acknowledge this potential scenario,” Ball commented. He also suggested that the Trump administration (with which he previously collaborated) will eventually recognize the necessity to “create substantial regulatory risk regarding the use of open-weight Chinese models.”

“You don’t need to ‘ban open source’ (a prevalent yet misguided argument in AI policy discussions),” Ball stressed. “You simply need to direct every agency to implement soft laws that promote FUD [fear, uncertainty, and doubt]. ‘A Federal Reserve Advisory Bulletin hinted that there may be backdoors in Chinese AI models.’ The reasoning doesn’t need to be solid; you just need to generate enough regulatory risk that every regulated entity withdraws.”

In contrast, Shakeel Hashim, editor of the AI-focused publication Transformer, argued that much of the concern is overstated, asserting that Kimi “likely does not possess dangerous cyber capabilities,” and the Chinese government would encounter “remarkably similar incentives” to regulate open Chinese models once such capabilities are developed.