The Rise of US-China AI Arms Control Negotiation Discourse: Situation Analysis and Response Strategies

Executive Summary

At the September Washington summit, the United States and China agreed to establish a mutual notification channel for AI incidents and a new 'Superintelligence Dialogue.' However, this is merely a crisis management gesture and can hardly be viewed as a substantive arms control regime. Immediately after the summit, President Trump stated that the US would not share AI governance with China, confirming a policy stance of 'achieving superintelligence' and 'winner-take-all' as official policy. Meanwhile, China maintains that sharing the authority to define safety and applying identical standards to US companies are prerequisites, making it structurally difficult to find common ground. Given the internal US structure where the White House's cooperation track and the Congress/regulatory agencies' pressure track operate separately, symbolic agreements between the leaders are highly unlikely to prevent the continuation of individual regulations, such as semiconductor and AI export controls. The most likely path is the baseline scenario (55–60% probability) where the dialogue is maintained nominally while substantive competition continues. South Korea must simultaneously establish a dual response system that separates security-linked sectors from commercial domains, secure a voice in multilateral norm-setting tracks, and preemptively review supply chain diversification. In particular, hastily expanding investments or contracts with China based solely on friendly signals from the White House carries high risks, necessitating a dual monitoring system that separately tracks changes in US Department of Commerce regulations.

I. Situation Analysis of the Issue

The Rise of US-China AI Arms Control Negotiation Discourse: Issue Situation Analysis

1. Background and Progress of the Issue

The Trump-Xi summit held in Washington on September 24 once again revealed the nature of bilateral relations surrounding AI. Immediately after the summit, the White House announced that the two countries agreed to establish a communication channel to mutually notify each other in the event of an AI incident [7][15]. It was agreed to hold the next meeting by November [4][7]. The establishment of this channel is not the first agreement reached this year. At the Beijing summit last May, the two leaders had already agreed to initiate an intergovernmental dialogue on AI [3][9]. The September Washington summit serves as a follow-up to that agreement [9].

A notable aspect of the summit was the war of words over terminology. The White House announced that it would use the term 'superintelligence (SI)' instead of 'AI' going forward [4]. President Trump used the term 'superintelligence' as political rhetoric that emphasizes development over control [4]. The Chinese side has never officially used this term [4]. This discrepancy in terminology aligns with previous EAI analyses showing that the two countries attach different political objectives to the term 'AI safety' [9]. The Trump administration addresses safety discussions only within the framework of maintaining a competitive advantage over China, while China has demanded sharing the authority to define safety and applying identical standards to US companies as prerequisites [9].

During the summit, President Xi Jinping mentioned that AI must not escape human control [6]. President Trump stated that they agreed new regulations were unnecessary [6]. These remarks are evaluated as "closer to gestures for domestic political consumption on both sides rather than normative achievements with substantive binding force" [6]. This interpretation is supported by the fact that the establishment of an 'AI hotline' for mutual notification of AI incidents had already been discussed in New York two days prior to the summit [6].

2. Current Situation

Less than a week after the summit, President Trump's stance became even clearer. During a meeting with AI industry executives at the White House on September 29, President Trump stated that the US would not engage in joint AI governance with China [12]. The reason was that it would make it difficult for US companies to stay ahead [12][14]. He noted that while he discussed AI during the summit with President Xi, they "did not discuss it much" and added, "I don't want to do anything" [12]. At the launch event for the America.gov portal, he remarked, "Whoever achieves superintelligence wins," adding, "There will be winners and losers, and there probably won't be a second place" [14]. He made it clear that the US currently leads China and all other nations, and intends to maintain this gap [14].

These remarks came amidst a series of hacking incidents. Despite hacking cases involving 'rogue AI' in several countries, including Australia, and warnings from industry leaders, President Trump continued to resist pressure for technology regulation [12].

China's response was not as public as that of the US. After the summit, President Xi Jinping offered only a restrained comment, stating that they had agreed on "strategic stability based on respect, fairness, and reciprocity" [10]. Beijing and Washington agreed to expand AI safety dialogue, increase communication between their militaries, and reduce tariffs on approximately $30 billion worth of 'non-sensitive' items [10]. Commentaries from Chinese state-run media outlets maintain a stance of shifting safety discussions toward the issue of global rule-making authority [9].

Silicon Valley's perspective is also skeptical. Skepticism has been raised within Silicon Valley regarding the hotline with Beijing proposed by Washington [17]. At the state dinner, Nvidia's Jensen Huang, AMD's Lisa Su, Elon Musk, and Apple's Tim Cook sat at the same table as Trump and Xi [17]. The very placement of industry leaders at the forefront of summit diplomacy demonstrates that the AI safety dialogue is inseparable from corporate interests.

3. Key Actors and Positions

The Trump Administrationaccepts AI safety cooperation only within the framework of maintaining a competitive advantage. President Trump himself repeatedly argues that joint governance with China would hinder US companies from securing technological dominance [12][14]. He is also passive about introducing new regulations and has directly rejected calls from the industry to slow down development [6][9]. However, he did agree to establish a hotline for mutual notification in the event of AI incidents, which is merely a crisis management channel and does not signify an expansion of commercial AI cooperation [6].

The Chinese Governmentseeks to utilize safety discussions as a means to secure global rule-making authority [6][9]. While rejecting US-led demands to slow down, Beijing exhibits a dual attitude by internally raising its awareness of AI risks [3]. President Xi Jinping's remarks on 'human control' strongly bear the character of a message for domestic political consumption [6].

Former NSC Official (Chris McGuire)is a key proponent of the discourse on AI arms control negotiations. He managed US-China AI policy at the NSC during the Biden administration, and prior to that, served as the State Department's senior expert on US-Russia arms control policy during the first Trump administration [2]. Drawing on this background and citing warnings from leading US AI experts, he is urging AI safety negotiations between Trump and Xi [2]. In an attempt to apply the Cold War experience of nuclear arms control to the AI sector, proposals have been raised to ultimately shape the regulatory framework into legislation embedded within a US-China agreement [8].

Silicon Valley Big Techfaces conflicting interests, particularly among companies invited to the summit's state dinner, such as Nvidia, AMD, and Apple. While they desire access to the Chinese market and the easing of export controls, they maintain a skeptical view of official cooperation channels, such as the hotline with Beijing [17].

The White House Cooperation Track and the Congress/Regulatory Pressure Trackoperate separately [3][6][9]. Analyses consistently suggest that even if symbolic agreements are reached between the leaders, individual regulations such as semiconductor and AI export controls are highly likely to continue independently [3][6][9].

4. Key Issues

The first issue is the political asymmetry of the term 'safety.' The US appropriates safety as a means to maintain its competitive advantage, whereas China appropriates it as a means to secure global rule-making authority [9]. Unless this asymmetry is resolved, procedural agreements like hotlines may accumulate, but they are unlikely to lead to substantive normative agreements.

The second issue is the gap between negotiation proposals and actual policy. While arms control proposals by former officials, including McGuire, possess concrete details [2][8], President Trump himself explicitly rejected joint governance with China immediately after the summit [12][14]. This creates a structure where the negotiation discourse of policy elites diverges from the actual choices made by the leader in the White House.

The third issue is the structural red line of the Taiwan question. Concerns have been raised that the AI safety dialogue could be intertwined with this red line and treated merely as a side deliverable of the summit [3][9]. This is a structure where discussions addressing the risks of military AI weaponization cannot be separated from the more fundamental security conflict of cross-strait relations.

The fourth issue is the effectiveness of the hotline. Although the establishment of an AI incident notification channel was announced [7][15][16], skepticism exists even within Silicon Valley regarding whether it possesses the foundation of trust required to function in an actual crisis [17]. While the procedure has been created, there is absolutely no operational experience.

II. In-Depth Analysis of the Issue

The Rise of US-China AI Arms Control Negotiation Discourse: In-Depth Issue Analysis

1. Analysis of Root Causes

At the root of President Trump's refusal to negotiate AI arms control lies a perception of hegemony that goes beyond mere negotiating strategy. At the launch event for the America.gov portal, he stated, "Whoever achieves superintelligence wins" [14]. He added, "There will be winners and losers, and there probably won't be a second place" [14]. This remark reveals a perception that views AI capability as a zero-sum asset, akin to military or economic power. In the nuclear arms race, the balance of terror known as mutually assured destruction (MAD) served as an incentive for arms control negotiations. In the AI competition, however, the perception of a "winner-take-all" outcome structurally removes the incentive to negotiate [14].

China operates under a different calculus. According to a CFR analysis, China seeks to utilize the US-led AI safety discussions as a means to secure global rule-making authority [9]. President Xi Jinping's remark that "AI must not escape human control" [6] is closer to a strategic move to secure China's voice in the process of shaping international norms. Commentaries from Chinese state-run media outlets have presented sharing the authority to define safety and applying identical standards to US companies as prerequisites [9]. This is a reaction against the double standard where the US applies loose standards to its own companies while imposing strict export controls on Chinese firms.

The terminology used by both sides is also a fundamental cause. The White House announced that it would use the term 'superintelligence (SI)' instead of 'AI' [4]. This expression is President Trump's political rhetoric, which emphasizes development over control [4]. China has never officially used this term [4]. The inability to agree even on the name of the technology that is the subject of negotiations is merely a surface symptom of the fact that the two countries attach different political objectives to the word "safety" [9].

2. Structural Context

Political Structure: Two Separate Tracks

The White House-led cooperation track and the Congress/regulatory-led pressure track already operate separately [9][6]. This is a structure where symbolic agreements between the leaders cannot prevent the continuation of individual regulations, such as semiconductor and AI export controls [9]. President Trump directly stated that although they discussed AI at the summit, they "did not discuss it much" [12]. He also remarked, "I don't want to do anything" [12]. This demonstrates that symbolic gestures at the level of summit diplomacy and policy decisions at the working level move on entirely different tracks from the outset. Another CFR analysis projects that this structure could ultimately result in "a US domestic regulatory regime embedded within a US-China agreement, and perhaps eventually legislation," but notes that it remains to be seen whether China would be open to dialogue within such a framework [8].

Economic Structure: Interests of Big Tech

The commercial interests of Silicon Valley Big Tech companies are deeply intertwined with AI safety discussions. At the White House state dinner, Nvidia CEO Jensen Huang and AMD CEO Lisa Su sat at the same table as Trump and Xi [17]. Elon Musk and Apple CEO Tim Cook were also present [17]. The hotline proposed by Washington to Beijing is facing skepticism even within Silicon Valley [17]. US companies harbor concerns that a cooperative framework with China could constrain their technological edge. President Trump's refusal to cooperate on AI governance with China, citing the reason that "it would make it difficult for US companies to stay ahead" [12][14], is closely aligned with these industry interests.

Security Structure: The Red Line of the Taiwan Question

The AI safety dialogue cannot be viewed in isolation from the structural red line of the Taiwan question [9][3]. Unless military tensions surrounding Taiwan are resolved, military confidence-building measures related to AI are unlikely to achieve a status beyond that of a side deliverable [9]. Immediately after the summit, Beijing and Washington also agreed to expand military-to-military communication [10]. However, much like the AI safety dialogue, this is closer to a minimal safety mechanism to prevent accidental clashes in the event of a crisis, rather than an agreement that alters the fundamental strategic competition.

4. Historical Precedents and Comparative Cases

Chris McGuire, the former White House National Security Council official leading this discussion, served as the State Department's senior expert on US-Russia arms control policy during the first Trump administration before managing US-China AI policy under the Biden administration [2]. His attempt to apply the Cold War experience of US-Soviet arms control to the AI sector highlights both the success factors and the limitations of nuclear arms control.

An Economic Times analysis points out that the logic of the current US-China AI competition resembles that of the Cold War nuclear arms race. It is a pattern where both countries develop increasingly destructive capabilities while simultaneously establishing limited crisis management mechanisms [1]. This similarity is supported by the fact that while Trump and Xi agreed to communicate in the event of an AI crisis, President Trump refused deeper risk-related cooperation [1].

However, there is a decisive difference between nuclear arms control and AI arms control. Nuclear weapons are held by a limited number of states, their testing and deployment can be verified via satellites, and their destructive power upon use is immediate and clear. In contrast, AI has blurred boundaries between civilian and military uses, its algorithms and model weights are virtually impossible to verify, and its damage often manifests cumulatively or indirectly. Hacking incidents involving "rogue AI" in several countries, including Australia [12], represent one facet of these asymmetric risks. This is why it is difficult to directly apply the frameworks of verifiability and deterrence envisioned by Cold War arms control to AI.

A comparison with the 2023 San Francisco APEC summit is also valid. While that summit was characterized by patching up relations strained by the spy balloon incident [9], the Washington summit was an occasion to emphasize the continuity of relationship management, building on the preceding summit diplomacy of the May visit to Beijing [9]. However, the substance of the outcomes is not significantly different. Both summits culminated merely in procedural achievements, such as the establishment of crisis management channels, while the fundamental structure of technological hegemony competition remained intact.

4. Key Variables Shaping Future Developments

The first variable is the outcome of the next dialogue scheduled for November [4][7]. Whether the next meeting of the 'Superintelligence Dialogue' addresses substantive agendas or remains a symbolic contact is expected to serve as a touchstone for gauging the nature of bilateral AI relations over the coming year.

The second variable is the intensity of regulatory pressure within the United States. As warnings from industry leaders and hacking incidents accumulate [12], Congress and regulatory agencies may attempt to check the White House's line of refusing cooperation. However, since President Trump reaffirmed his stance of avoiding regulation at the America.gov event, calling it "the only way to maintain this gap" [14], the likelihood of a policy shift within the administration appears low at present.

The third variable is the change in China's internal perception of risk. While officially rejecting US-led demands to slow down, China has exhibited a dual attitude by internally raising its awareness of AI risks [3]. Whether President Xi Jinping's remarks on "human control" [6] remain a rhetorical gesture or lead to actual strengthening of domestic AI governance will determine the credibility of Beijing's negotiations.

The fourth variable is the situation in Taiwan. If military tensions surrounding Taiwan escalate, there is a risk that even the AI incident notification channel currently being established could cease to function. Conversely, if tensions are managed at a controllable level, there is scope for the AI hotline to serve as a minimal crisis management infrastructure between the two countries.