Kimi K3 sparks debate in Silicon Valley! Anthropic CEO's first response: Never advocated banning open source AI models, but requested testing

The Zhitong Finance App learned that a “civil war” in Silicon Valley caused by the Chinese AI company Moonshot AI (Moonshot AI) released an open source model, Kimi K3, is tearing apart the US tech community. From Anthropic CEO Dario Amodei to Nvidia CEO Huang Renxun, from OpenAI to Microsoft, Silicon Valley leaders have had a fierce open battle on how Washington should respond to China's AI breakthroughs.

The core of this debate is no longer simply a “technology gap,” but rather whether the US should curb the rise of AI in China by banning the open source model — and how this choice will reshape the power pattern of the global AI industry, touching on the fundamental conflict between America's AI leadership, national security, and industrial interests.

Anthropic's latest response: never advocated a ban but called for mandatory testing

Anthropic was in the most embarrassing position in the whole incident. As a victim of being accused of being “stolen,” the closed-source AI company was fiercely criticized by peers for not signing a joint letter supporting open source.

Anthropic CEO Dario Amodei finally broke his silence on July 28 and published a blog post entitled “Our Position on the Open Weighting Model” in the face of outside accusations that it “pushes to ban open source models to protect commercial interests.” He made it clear: “I want to be clear that Anthropic has never advocated a ban on open weighting models.” Those open weight models without dangerous features are “a kind of public asset.”

But he also drew a red line. According to Amodei, an open weighted model with no dangerous features is a public asset that can bring value to businesses, developers, and researchers. But at the same time, he stressed that the protectionist ban cannot resolve the national security issues he is most concerned about. He advocates a severe crackdown on “industrial-scale distillation operations” and requires that both open source and closed source models undergo mandatory safety testing before release.

However, the reason Amodei declined to sign the joint letter was that he did not agree with the letter's view that “an open weighting model is bound to be easier to implement protective measures.” In his opinion, “the opposite may be the case”. He used biological aggression as an example of asymmetry — powerful models can quickly create mass destruction viruses, while defense takes years.

Kimi K3: An “open source engine” that is tearing Silicon Valley apart

In mid-July, the Chinese AI startup Moon's Dark Side released the open source weight model Kimi K3 — a giant model with 2.8 trillion parameters. The model performed well in coding benchmarks, performing close to America's most advanced AI systems, but at a much lower cost. The media called this incident the “Sputnik moment” for open source AI.

Kimi K3 is no ordinary model iteration. It scored 57 points on the Artificial Analysis Intelligence Index, ranking third — behind Anthropic's Fable 5 (60 points) and OpenAI's GPT-5.6 Sol (59 points), surpassing Claude Opus 4.8 (56 points). In terms of front-end coding capabilities, K3 reached the top of the Arena rankings within 24 hours of release.

What really sparks anxiety in Silicon Valley is the simultaneous occurrence of “performance+price+openness”: K3's API price is less than one-third of Fable 5, and the full weight will be publicly released on July 27. More importantly, before the K3, the Kimi series model was quietly embedded in the Silicon Valley production chain — the code tool Cursor, which SpaceX is buying for about 60 billion US dollars, and its core runs on Kimi K2.5.

From closed source bastions to open source coalition factions

The release of Kimi K3 has heated up the long-standing dispute between open source and closed source routes in Silicon Valley.

One side is a closed source faction led by Anthropic and OpenAI, which believes that some AI models are too dangerous, should not be developed in an open manner, and must be strictly controlled by companies like them to ensure safety. The other side is an open camp led by Nvidia, Microsoft, and Meta, which argues that the open source AI model must remain open so that people can further develop technology and establish new businesses.

On July 24, Hwang In-hoon launched a joint letter — “Open Weights and American AI Leadership” in his first X platform post, calling for the development of the Open Weights AI model. In his post, he wrote, “AI will transform every industry, empower every company, and be built together by every country. The open source model strengthens security, accelerates innovation and technology adoption, and enables sovereignty. The world needs both a cutting-edge closed source model and a cutting-edge open source model.”

The letter, which was initially signed by 25 organizations, quickly doubled to 50 within a day, with additional signatories including OpenAI and Google. Microsoft CEO Satya Nadella later posted that open source software “is critical to a healthy AI ecosystem.” Elon Musk retweeted and wrote, “Wong In-hoon is right, I fully support it.” Meta CEO Zuckerberg also said, “Open source is a positive and important force.” Google CEO Sundar Pichai also joined the supporters on Saturday.

“Sputnik moment” sparks panic in Washington

The launch of Kimi K3 also caused a sharp shock in Washington. Michael Kratsios, director of the White House Office of Science and Technology Policy, publicly accused The Dark Side of the Moon of stealing Anthropic's Claude Fable 5 model through “distillation” technology during the development of Kimi K3. He claims the company built a “complex internal platform” for large-scale distillation and quickly switched multiple access methods to evade detection. Kratsios also alleges that Dark Side of the Moon obtained a server equipped with the Nvidia GB300 in violation of US export controls.

US Treasury Secretary Scott Bessent immediately warned that if the accusations are true, Chinese companies may face financial sanctions or be included in the Commerce Department's list of entities, restricting their access to US technology. “We have the power to sanction them,” Bessent said.

However, these allegations were immediately challenged. White House insiders are skeptical about the “distillation” accusation: “Distillation doesn't make sense at the core.” The global AI research community also generally believes that distillation does not constitute intellectual property theft, and points out that the US side lacks evidence.

White House Internal Line Dispute: Bezent's “Big Stick” vs. Lutnick's “Scalpel”

The Trump administration is also deeply divided on how to deal with the rise of AI in China. Hardliners, such as Bessent and Kratsios, advocate protecting US intellectual property rights and adopting sanctions and export control measures. Bessent threatened to add Chinese AI companies to the list of entities; Kratsios directly named and accused the dark side of the Moon on the X platform.

Moderates, represented by Commerce Secretary Howard Lutnick, are more cautious, fearing that excessive restrictions will damage American companies' access to cheap and efficient technology.

Silicon Valley startups have expressed their positions with actions. On July 23, the “Little Tech Association” (Little Tech Association) representing nearly 200 Silicon Valley startups sent a letter to the Trump administration warning that “hundreds of US companies will go out of business in an instant” if China's open source model is banned. The Association's executive director Harry Godfrey called on policymakers to use a “scalpel rather than a sledgehammer.” The joint letter stated that in order for the US to maintain its leading position in AI, it needs to do two things: have a world-leading US open weighting model, and allow US developers to continue to use the global open Chinese model.

Meanwhile, Wong In-hoon will meet with Mark Warner, Democratic Party Chief Member of the Senate Intelligence Committee, in Washington this Tuesday to discuss AI security and open source issues. OpenAI CEO Sam Altman will also meet Bessent and Lutnick in Washington. This intensive series of events on Capitol Hill comes as the Trump administration prepares to release a proposed security review framework for AI models.

Industry impact: the ripple effect from chips to cloud computing

The impact of this debate is spreading across the AI industry chain. For chipmakers, Nvidia is the biggest beneficiary of the open model — the open source model requires more GPUs for training and reasoning. Hwang In-hoon's public stand was not just ideological, but based on clear commercial interests.

When it comes to cloud service providers, Microsoft and Amazon are equally clear: no matter what model a company chooses, it ultimately needs to run in the cloud, so the more open models, the greater the demand for cloud services.

For startups, the joint letter warns that restricting open source software could solidify OpenAI and Anthropic's leading position in the AI field and unfairly stifle the ability of other companies to develop competitive products. Silicon Valley venture capitalist Bill Gurley said, “One faction wants OpenAI and AnthroPic to control everything, while the other group wants everyone else, including customers.”

Regarding China's AI industry, Beijing has clearly warned that it will take reciprocal measures. China's Ministry of Commerce accuses the US of engaging in “AI hegemonism” and threatens to take retaliatory measures if the US imposes sanctions. Analysts warned that US accusations and restrictions could disrupt the US-China AI dialogue originally scheduled to be held in September.