US Review of Moonshot AI Sanctions and Structural Shift in US-China AI Hegemonic Competition: Strategic Response Measures for South Korea

Executive Summary

The US review of sanctions against Moonshot AI is not merely an intellectual property dispute. It signals a structural shift triggered by the fact that the premise of the US containment strategy, which equates 'computing power superiority with AI performance superiority,' has been directly challenged by the release of Kimi K3. While US technology sanctions against China are expanding their scope from hardware semiconductor controls to blocking access to AI software models and further to intellectual property disputes, the spread of open-weight models and China's algorithmic innovation capabilities suggest that the sanctions may be more counterproductive than effective. South Korea, by virtually dominating the core AI hardware supply chain, including HBM, is in a position of structurally enhanced negotiation leverage. However, it also faces a dual vulnerability of being simultaneously pressured to choose sides between the US's unilateral access restrictions and its economic ties with China. Therefore, South Korea must adopt a 'strategic dual positioning' as its core response strategy, deepening its strategic indispensability within the US AI ecosystem while concurrently developing its own AI capabilities and alternative ecosystems, thereby maintaining a structural foundation for its negotiating power under any scenario.

I. Issue Situation Analysis

US Review of Moonshot AI Sanctions: Allegations of IP Theft and Export Control Violations

Issue Situation Analysis

1. Background and Progression of the Issue

Artificial intelligence has rapidly emerged not just as a domain of technological innovation but as a strategic asset that dictates national security, military power, and economic hegemony. The United States and China are the two main poles of this competition, with both nations recognizing AI as a decisive tool for hegemonic competition and prioritizing it at the highest level of their national strategies [2]. Within this structural competitive context, following the shockwaves sent through Silicon Valley by the Chinese startup DeepSeek's low-cost, high-efficiency model in early 2025, the release of Moonshot AI's Kimi K3 in July 2026 is being evaluated as another structural turning point in the US-China AI competition.

Historically, when a dominant power in a hegemonic competition perceives a decline and recognizes the threat from a rising competitor, attempts to block the competitor's economic growth by cutting off supply chain access have been repeated. The US prohibition on the sale of advanced satellite technology to Japan in the 1980s and its control over semiconductor exports to China in the late 2010s are extensions of this historical pattern. China has responded by advancing its technological base, and the emergence of Kimi K3 can be seen as the latest example of this response.

The US government has expanded the scope of export controls from hardware semiconductors and equipment to access to frontier AI models to prevent security-concerned nations like China and Russia from potentially misusing advanced AI systems [5]. In June 2025, the Trump administration took measures to retroactively enforce restrictions on non-US persons' access to Anthropic's top-performing model, Fable (Claude Fable 5), a move considered a signal of the structural shift in the US export control system, expanding comprehensively beyond hardware into the realm of AI software models [5]. Amidst this trend, in April 2026, the US Department of State began seeking international cooperation by disseminating information through diplomatic channels to allies, alleging that Chinese companies, including Moonshot AI, were extensively stealing intellectual property from US AI research institutions [12].

2. Current Situation

Kimi K3, released by Moonshot AI on July 15, 2026, is an open-weight model with 2.8 trillion parameters. It has shocked the global tech industry by demonstrating performance close to Anthropic's Claude Fable 5 and OpenAI's ChatGPT-5.6 on complex tasks such as coding and reasoning [4][14]. Particularly noteworthy is that this model achieved performance close to top US AI models at 30-50% lower cost through algorithmic optimization and model architecture innovation, despite US semiconductor export controls on hardware. This is evaluated not as a mere one-off incident but as a signal heralding a structural shift in the US-China AI competition [2].

On July 22, immediately after the release of Kimi K3, Michael Kratsios, Director of the White House Office of Science and Technology Policy (OSTP), officially claimed on social media that Moonshot AI had extensively distilled Anthropic's Fable model to develop Kimi K3 [12]. He strongly criticized this as "a massive covert industrial distillation operation to steal America's proprietary technology and undermine American research" [12]. Furthermore, he raised suspicions of violating US export control regulations, suggesting that Moonshot AI likely obtained servers equipped with Nvidia's GB300 chips, which are prohibited export items, in Thailand and used these chips to train its AI models [7].

In response, global AI experts immediately refuted these claims. Numerous researchers pointed out the lack of concrete evidence for the distillation allegations and emphasized that AI model outputs themselves are not subject to copyright protection. Some experts questioned the credibility of the US claims by arguing that it is technically implausible to train a completely new frontier model in just 15 days, given that Fable was released on July 1 and K3 on July 15 [10]. These counterarguments support the view that the current sanctions review is driven more by geopolitical competition logic than by technical facts.

Internal conflicts within the Trump administration are also surfacing. A hardline faction, centered around Kratsios, advocates for immediate and broad sanctions against Chinese AI companies, while elements with industry-friendly stances express concern about the negative ripple effects that hasty regulations on open-weight models could have on the entire US AI industry [2]. Over 20 companies, including Nvidia, have officially expressed opposition to broad regulations on open-weight models [2].

3. Key Actors and Their Positions and Interests

Hardline faction within the US Trump administrationviews this incident as part of China's systematic IP theft strategy and insists on establishing a precedent through strong sanctions. OSTP Director Kratsios has publicly hinted at the possibility of sanctions against Moonshot AI, serving as a warning message not only to one company but to Chinese AI startups in general [14]. The US Department of State has also been attempting to build a multilateral cooperation framework since April by sharing details of Chinese companies' IP theft with allies [12].

US AI Industry and Nvidiaare complexly divided in their interests. While Anthropic is a victim whose technology has allegedly been used without authorization, hardware and infrastructure companies, including Nvidia, are concerned that broad regulations on open-weight models could weaken their global competitiveness [2]. The opposition expressed by these over 20 companies symbolically highlights the internal industrial and political conflict within the US regarding the effectiveness and scope of sanctions.

China's Moonshot AIofficially denies the allegations and has secured a degree of support from the global AI expert community. Moonshot AI's assertion that the technical achievements of Kimi K3 are the result of independent innovation, coupled with the refutations from international researchers pointing to the lack of evidence for distillation claims, highlights the political nature of the US's assertions [10]. The Chinese government is also expected to strengthen its commitment to protecting its domestic AI companies, framing this incident as unfair pressure from the US to maintain its technological hegemony.

Global AI Expert Communitycriticizes the US government's claims as politically motivated and emphasizes the lack of technical grounds [10]. Their refutations are escalating into issues of credibility for US AI governance, with CFR even offering a sharp assessment that the US is losing the AI credibility war to itself [6].

South Korean Companies and Governmentare not direct parties to this incident but are among the actors with the most complex structural interests. As demonstrated by the case where the order to block non-US persons' access to Anthropic's model was issued just five days after SK Telecom and Nvidia announced their gigawatt-class AI cloud partnership, the fact that US export control decisions are primarily driven by self-centered national security and industrial interests rather than alliance cooperation logic has starkly exposed the structural vulnerabilities of South Korean companies [5]. Simultaneously, the leading position of Samsung Electronics and SK Hynix in HBM serves as a factor that structurally strengthens South Korea's negotiation leverage [11].

4. Summary of Key Issues

The key issues in this incident can be summarized into four main dimensions.

First, the issue of the technical reality and legal basis of the distillation allegations. The US government claims that Moonshot AI developed Kimi K3 by illicitly distilling Anthropic's Fable model, but international legal standards regarding the copyrightability of AI model outputs and the illegality of distillation are not yet established. As experts point out, the question of whether large-scale distillation is technically feasible within the extremely short period of 15 days remains unresolved [10].

Second, the issue of the effectiveness versus counterproductivity of export controls. "US export controls have been incrementally imposed as a makeshift measure, blending security imperatives with industrial-political motives. The case where a Chinese startup, Z.ai, released a model with similar performance without any restrictions immediately after the ban on non-US persons' access to Anthropic's top-performing model was implemented, symbolically demonstrates that the current control system fails to substantially curb the development of China's AI capabilities while paradoxically weakening the global competitiveness of US companies." [8] Kimi K3 is also an extension of this paradox, marking a case that directly refutes the core premise of the US containment strategy, 'computing power superiority equals AI performance superiority' [2].

Third, the issue of the scope and limitations of regulating open-weight models. Due to their nature, open-weight models are virtually impossible to control once released. Broad regulations could undermine the openness and innovative momentum of the US AI ecosystem. The backlash from the industry, including Nvidia, clearly illustrates this dilemma [2].

Fourth, the issue of the geopolitical ripple effects of sanctions. If this sanctions review expands beyond the AI field to encompass the broader US-China competition in the technology and IT sectors, allies, including South Korea, will face increasingly acute pressure to choose between their dependence on the US AI ecosystem and their economic ties with China. "As the bottleneck in the AI supply chain expands beyond GPUs to core hardware value chains such as memory, substrates, and optical components, which South Korea virtually dominates, South Korea's negotiation leverage is structurally strengthening." [5] This also presents a structural opportunity for South Korea to position itself not as a passive victim but as an active strategic actor.

II. In-depth Issue Analysis

US Review of Moonshot AI Sanctions: Allegations of IP Theft and Export Control Violations

In-depth Issue Analysis

1. Analysis of the Root Causes of the Issue

The root cause of this incident goes beyond mere allegations of intellectual property infringement; it is where the structural contradictions of the US-China AI hegemonic competition have simultaneously surfaced. The US has pursued a strategy of suppressing China's AI capability development using semiconductor export controls as a key tool. However, the emergence of Kimi K3 directly refuted the core premise of this strategy, namely the equation 'computing power superiority equals AI performance superiority,' by demonstrating that this equation may no longer hold true [2]. In other words, the US containment strategy stemmed from a structural misjudgment that underestimated China's algorithmic innovation capabilities, and the current sanctions review is strongly characterized as a political reaction to the exposure of that misjudgment.

More deeply, this incident raises fundamental questions about whether 'access control' can function as an effective deterrent in AI technology competition. In June 2025, the US took measures to retroactively enforce restrictions on non-US persons' access to Anthropic's Fable model [5], expanding the regulatory front from hardware controls to software model controls. However, the case where the Chinese startup Z.ai released a model with similar performance without any restrictions immediately after this measure was implemented symbolically demonstrates that the current control system paradoxically weakens the global competitiveness of US companies while failing to substantially curb the development of China's AI capabilities [8]. The Kimi K3 incident is a case where this structural paradox has been reconfirmed, and the US government's decision to resort to sanctions, regardless of the sufficiency of evidence, can be interpreted as an attempt to find a political breakthrough for this paradox.

Furthermore, another root cause of this incident is that the spread of open-weight models is shaking the very logical foundation of the existing export control system. Open-weight models have their weights disclosed, making it difficult to selectively block access by specific companies or countries. Once released, these models become virtually uncontrollable. The opposition expressed by over 20 companies, including Nvidia, against hasty regulations on open-weight models reflects the industry's fundamental skepticism about the effectiveness of such regulations [4]. The gap between technological realities and regulatory logic thus transforms this incident from a mere enforcement issue into a fundamental redesign problem for AI governance.

2. Structural Context

Political Structure

At the political level, this incident is also an externalization of the line conflict within the Trump administration. The hard-line sanctions faction, represented by Office Director of the White House Office of Science and Technology Policy Kratios, advocates for immediate and broad sanctions against Chinese AI companies, while the industry-friendly faction counters with the argument that comprehensive regulation of open-weight models could stifle the innovation ecosystem of the US AI industry itself [4]. This conflict goes beyond a mere difference in policy lines, reflecting a fundamental strategic debate about how the United States will maintain its AI hegemony. In other words, it is a question of whether to maintain superiority by containing China or to widen the gap by accelerating the innovation speed of American companies.

At the international political level, it is noteworthy that the United States began seeking multilateral cooperation by sharing the realities of intellectual property theft by Chinese AI companies with its allies through diplomatic cables in April 2026 [12]. This is an attempt by the United States to incorporate its allies into the regulatory framework to enhance the effectiveness of export controls, demonstrating that AI governance issues are expanding beyond bilateral relations into a field of multilateral geopolitical competition. However, at the same time, this multilateral approach creates a structure that effectively forces allies to participate in a US-centric technology bloc, thereby acting as a new source of tension in alliance management [11].

Economic Structure

At the economic level, this incident shows that the supply chain structure of the AI industry is emerging as a key battlefield for geopolitical weaponization. Modern great powers compete not with bullets and bombs, but through global supply chains, utilizing economic means such as trade, finance, manufacturing, and access control to technology as strategic weapons. In the field of AI, this competition is unfolding around access to GPU chips, High Bandwidth Memory (HBM), data center infrastructure, and frontier AI models themselves [11]. The US export control on Nvidia's GB300 chips is the hardware front of this competition, and the blocking of access to the Fable model signifies the opening of the software front.

However, this economic structure also imposes significant costs on the United States. The bottlenecks in the AI supply chain are spreading beyond GPUs to the entire core hardware value chain, which is virtually dominated by South Korea and Taiwan, including memory, substrates, and optical components [5], indicating that the US's unilateral control capability has structural limitations. Furthermore, regulating open-weight models restricts the global market access of US AI companies, weakening their revenue base, which in turn can lead to a vicious cycle of reduced R&D investment capacity. The assessment that "the current export control system is paradoxically weakening the global competitiveness of US companies while failing to substantially curb the development of China's AI capabilities" [8] accurately captures this structural dilemma.

Security Structure

At the security level, this incident shows that the US concern about the military application of AI technology is the core driver for expanding regulations. The US government has continuously expanded the scope of export controls to block the possibility of security-concerned countries like China and Russia misusing advanced AI systems for military and intelligence purposes [5]. Large models with excellent coding and reasoning capabilities, such as Kimi K3, particularly stimulate the vigilance of security authorities due to their potential for military applications including cyber operations, autonomous weapon systems, and intelligence gathering and analysis [8].

However, there is a significant gap between the logic of security and the reality of technology. Once open-weight models are released, it is technically almost impossible to selectively block their military access by specific actors. This raises fundamental questions about the effectiveness of controlling open-weight models as a means of regulation to achieve security objectives, and it gives rise to criticism that the US government, while citing security logic, is actually motivated by industrial politics to delay the technological development of competitor nations [8].

3. Comparison of Historical Precedents and Similar Cases

Historically, attempts by dominant powers to block the supply chain access of rising competitor nations in hegemonic competition have repeatedly occurred, and their outcomes have generally unfolded in directions contrary to the expectations of the blockading country. When Britain blocked German shipping routes in the early stages of World War I, Germany responded by improving industrial production efficiency. When the US raised rubber prices for Japan, Japan switched to cheaper sources in Liberia. Japan and China responded to US technology controls by advancing their own technological foundations. This historical pattern consistently shows that while containment strategies cause friction in the short term, they paradoxically lead to the strengthening of the self-reliance capabilities of the contained country in the long term.

The case of US semiconductor export controls against China is the most directly comparable precedent to the current review of AI sanctions. Since 2018, as the US intensified export controls on Chinese semiconductor companies like Huawei and SMIC, China has set semiconductor self-sufficiency as its top national strategic priority and concentrated massive investments. As a result, while China's access to advanced semiconductors was limited in the short term, it accelerated the development of China's indigenous semiconductor industry in the medium to long term. The assessment that "the strengthening of US export controls has been cumulatively implemented on an ad hoc basis, with a mixture of security needs and industrial political motives" [8], suggests that this pattern is likely to be repeated in the AI field.

A more direct similar case is the DeepSeek incident in early 2025. DeepSeek developed low-cost, high-efficiency models despite the constraints of US semiconductor export controls, shocking Silicon Valley and cracking the fundamental premise of the US strategy that 'computing power superiority = AI performance superiority' for the first time [2]. Kimi K3 is an intensification and repetition of this pattern, supporting the core argument of the plan's intent that US sanctions and controls are paradoxically accelerating self-reliant technological development by aligning innovation incentives between Chinese companies and the government. Indeed, the case of Anthropic blocking non-US access to its Fable model, followed by the launch of a similar performance model by Chinese startup Z.ai without restrictions, serves as vivid evidence that containment is rather acting as a catalyst for accelerating China's indigenous development speed [8].

Meanwhile, the case of US sanctions on Japanese semiconductors also provides an important point of comparison. In the 1980s, the US felt threatened by the advancement of Japanese semiconductor companies and imposed trade pressure and restricted access to technology, which led Japanese companies to strengthen their independent technological capabilities in materials, components, and equipment. However, Japan's response was limited due to its security alliance with the US, whereas China has no such constraints, allowing for a much more aggressive self-reliance strategy, which represents a structural difference [11].

4. Key Variables in Issue Development

The key variables that will determine the future development of the issue can be summarized into four main points.

First, the sufficiency of evidence regarding the distillation allegationsis crucial. Global AI experts refute the US claim of distillation, arguing that there is insufficient evidence, and point out that the claim of training a completely new frontier model in just 15 days, from the release of the Fable model (July 1) to the launch of Kimi K3 (July 15), is technically implausible [10]. Furthermore, the legal argument that AI model outputs are not subject to copyright protection is also being raised [10]. If the US government fails to present legally valid evidence, the sanctions will struggle to gain international legitimacy and will face limitations in garnering cooperation from allies.

Second, the outcome of the line conflict within the Trump administrationwill determine the scope and intensity of sanctions. The conflict between the hard-line sanctions faction and the industry-friendly faction will determine the scope and intensity of sanctions. With over 20 companies, including Nvidia, opposing comprehensive regulation of open-weight models [4], industry lobbying pressure is expected to significantly influence policy decisions. Particularly, given the Trump administration's tendency to prioritize industrial interests, it is highly likely that the outcome will be in the form of targeted sanctions aimed at specific companies rather than broad open-weight regulations.

Third, the speed at which China secures its own computing poweris critical. As the analysis suggests, "if China secures sufficient additional computing power in the future, the US-China AI gap is likely to narrow even faster" [2], the speed at which China overcomes current hardware constraints is a key variable determining the medium- to long-term landscape of US-China AI competition. Allegations of Nvidia GB300 chip access via third countries like Thailand suggest that China is already partially overcoming hardware constraints through indirect routes, raising fundamental questions about the effectiveness of export controls [7].

Fourth, the possibility of establishing a multilateral export control systemis important. Export controls pursued unilaterally by the US have limitations in effectiveness, and it is difficult to block China's circumvention routes without the cooperation of allies. While the US is requesting cooperation from allies through diplomatic cables [12], the choices these allies make between their economic interests and security logic will determine the feasibility of a multilateral AI governance system. Amidst the high likelihood of a "fragmented competitive landscape where US export controls maintain partial effectiveness while China's indigenous development is simultaneously accelerated" [8], the establishment of a multilateral governance framework will act as a key variable in moderating the deepening speed of this fragmented competition.