The Rise of US-China AI Arms Control Negotiations: The Dual Structure of Establishing Dialogue and Rejecting Substantive Regulation
Executive Summary
At the Washington summit on September 24, the United States and China agreed to establish a dialogue on superintelligence (SI), but President Trump reaffirmed his stance against joint regulation with China immediately afterward. While the Chinese side does not reject safety discussions themselves, it has set the sharing of defining authority as a precondition, indicating that both sides prioritize relationship management over substantive binding power. This suggests that the AI safety dialogue is a subordinate agenda item tied to a trade truce package that includes approximately $30 billion in tariff cuts, demonstrating that industrial and security interests are being operated without separation within the framework of US-China strategic competition. South Korea should not overinterpret this dialogue as the starting point for substantive norm-making. Instead, it needs to establish a dual response system that separates security-linked technology controls from commercial domains, while simultaneously participating in multilateral norm tracks that are not subordinate to the bilateral US-China framework. Furthermore, South Korea should continuously track whether the AI and military hotlines are actually operational to review the necessity of introducing similar crisis management channels domestically.
I. Analysis of the Current Issue
The Rise of US-China AI Arms Control Negotiations: Analysis of the Current Issue
1. Background and Progress of the Issue
This is not the first time the United States and China have established an intergovernmental dialogue channel on AI. At the Trump-Xi summit held in Beijing last May, the two leaders had already agreed to initiate an intergovernmental dialogue on AI [3][9]. The summit held in Washington on September 24 serves as a follow-up to this agreement [3][9]. President Trump's personal disclosure that he had prepared a White House state dinner aligns with the symbolism of President Xi's first visit to the White House in 11 years [6][9].
Two days before the summit, discussions on establishing an "AI hotline" for mutual notification in the event of AI accidents were first disclosed in New York [9]. Following the summit, the White House announced that the two countries had established the "U.S.-China Superintelligence (SI) Dialogue" to exchange views on the risks and benefits of superintelligence [4]. The next meeting is scheduled to be held by November [4][7]. A noteworthy point is the choice of terminology. The Trump administration opted to use the term "superintelligence (SI)" instead of "AI," which is closer to political rhetoric emphasizing development supremacy over technology control [4]. The Chinese side has never officially used this term [4]. At the same meeting, agreements were also reached on expanding communication between the militaries of both countries and cutting tariffs on non-sensitive items worth approximately $30 billion [10].
Amid this trend, Chris McGuire, a former NSC official now with the Council on Foreign Relations (CFR), proposed a detailed design for a US-China AI safety agreement, drawing on his experience handling US-Russia arms control negotiations during the first Trump administration [2]. His proposal, which came amid accumulating warnings about AI risks from the industry, served as a catalyst for the "AI arms control" framework to emerge as an official agenda item in the United States [2][8].
2. Current Situation
Developments following the summit show a coexistence of symbolic achievements and substantive deadlock. While the White House self-evaluated the summit as "one of friendship, strength, and success" [10], President Trump made it clear at a separate event immediately after the summit that "whoever captures superintelligence will be the winner," emphasizing the maintenance of competitive advantage over cooperation with China [14]. At a meeting with AI executives at the White House on September 29, President Trump reaffirmed his rejection of pressure to regulate AI jointly with China [12]. He stated that he "did not discuss" the AI issue much with President Xi and added, "I don't want to do something" [12]. This remark came amid ongoing industry warnings and AI-related hacking incidents in places like Australia [12].
China's response has been relatively restrained. President Xi went no further than mentioning "strategic stability based on respect, fairness, and reciprocity" [10]. Analysts evaluate that China is adopting a dual approach, not rejecting safety discussions outright but refusing to comply with US-led demands to slow down development [3]. The South China Morning Post (SCMP) analyzed that while this AI dialogue channel is meaningful as an official channel for stabilizing bilateral relations, it is characterized more by relationship management than immediate trade achievements [18]. The CFR also characterized the summit as showing the "limits of photo-op summits," pointing out the gap between the actual level of agreement and the trend of groups concerned about the loss of human control over large AI models urging both the US and China to slow down development and establish safeguards [11].
The Economic Times likened the current phase to the Cold War nuclear arms race, assessing that while the two countries agreed to communicate in the event of a crisis, deeper cooperation for fundamental risk management was rejected [1]. The Peterson Institute for International Economics (PIIE) reported that expectations were not high even ahead of the summit, presenting an inertial outlook that there would be no major changes in either country's policies [5]. Nikkei Asia reported that the hotline proposed by Washington is facing skeptical reactions within Silicon Valley [17]. The fact that industry leaders such as Nvidia's Jensen Huang, AMD's Lisa Su, Elon Musk, and Apple's Tim Cook sat at the same table as Trump and Xi at the state dinner demonstrates that governments are not the only substantive stakeholders in safety discussions [17].
3. Key Actors and Positions
The Trump Administrationaccepts AI safety discussions only within the framework of maintaining a competitive advantage [9]. It adopts a dual approach, drawing the line at introducing new regulations while agreeing to establish communication channels for crisis management [6]. The attempt to rename "AI" as "superintelligence" is interpreted as a domestic political message prioritizing development speed over safety [4].
The Chinese Governmentadopts a dual stance, officially rejecting US-led demands to slow down while internally expanding its awareness of risks [3]. President Xi's reference to "strategic stability" is read as cautious rhetoric aimed at avoiding disruption to the basic framework of bilateral relations [10].
Former NSC Official (Chris McGuire)is the actor who proposed a concrete negotiation plan based on US-Russia arms control experience, triggering policy debates within the United States [2]. His proposal differs in tone from the official White House stance, demonstrating that tracks surrounding AI governance are separated even within Washington.
Industry Actors(such as Nvidia, AMD, and Apple) have positioned themselves as substantive stakeholders in summit diplomacy, as evidenced by the seating arrangement at the state dinner [17]. They occupy a contradictory position, remaining cautious about strengthening regulations while continuing to issue warnings about the risks of AI accidents [12].
Think Tanks such as the Carnegie Endowment for International Peace and the CFRraise the need to expand AI safety discussions beyond the bilateral US-China framework to multilateral norm tracks such as the UN, addressing who holds the authority to set AI guardrails as a core issue [13].
4. Summary of Key Issues
First, the very definition of the concept of "safety" remains unagreed upon between the two countries. The prevailing assessment is that the United States approaches safety as a means to maintain its competitive advantage, while China views it as a means to secure global rule-making authority [9]. Second, establishing hotlines and dialogue bodies for crisis management is an entirely different level of agreement from substantively slowing down development. Only the former has been achieved so far, while President Trump has explicitly rejected the latter [12][14]. Third, given the domestic US structure where the White House-led cooperation track and the Congress/regulatory agency-led pressure track operate separately, symbolic agreements between leaders are unlikely to lead to the relaxation of individual regulations, such as semiconductor and AI export controls [9]. Fourth, the gap between industry-led risk warnings and actual policy responses is widening. Even as AI-related hacking incidents in places like Australia are reported, the US administration leans toward prioritizing development supremacy over regulation [12]. A key point to watch is whether this gap will act as pressure to facilitate multilateral norm discussions outside of bilateral US-China negotiations in the future.
II. In-Depth Analysis of the Issue
The Rise of US-China AI Arms Control Negotiations: In-Depth Analysis of the Issue
1. Analysis of Root Causes
The root cause of the rise of AI arms control negotiation theories lies in the clash within the United States between the AI safety discourse and the discourse on maintaining a competitive advantage over China. Industry leaders and former government officials demand bilateral cooperation to control AI risks [2][8]. Conversely, President Trump perceives this as a weakening of competitiveness. He stated, "whoever captures superintelligence will be the winner" [14]. This remark demonstrates which side the United States prioritizes between cooperation and competition.
The weight of McGuire's proposal stems from his career background. He handled US-Russia arms control negotiations during the first Trump administration [2]. Subsequently, he led US-China AI policy at the NSC under the Biden administration [2]. The very attempt to apply nuclear arms control experience to the AI domain shows that US policy elites are elevating AI to the category of strategic weapons rather than conventional weapons. The Economic Times likened this trend to the "logic of the Cold War" [1]. It describes a structure where both sides develop increasingly destructive capabilities while simultaneously establishing limited crisis management mechanisms [1].
There are also instances where the risk of AI weaponization has actually materialized. AI-linked hacking incidents in places like Australia added credibility to industry warnings [12]. This transformed abstract security discourse into concrete accident cases, providing substantive momentum to discussions on establishing an AI hotline [9].
2. Structural Context
Political Structure: The White House's cooperation track and the Congress/regulatory agency's pressure track operate separately [3][9]. President Trump agreed to establish a dialogue body at the leadership level. At the same time, he draws the line at substantive regulatory strengthening [12][14]. This dual structure is not accidental. The Trump administration addresses AI simultaneously in both industrial and security policies, but intentionally operates the two domains separately. The scene at the state dinner where Nvidia CEO Jensen Huang and AMD CEO Lisa Su sat at the same table as Trump and Xi symbolizes this [17]. It is a structure where security dialogue and industrial interests are intermingled in a single space.
China's structural approach is different. President Xi offered only the restrained expression of "strategic stability based on respect, fairness, and reciprocity" [10]. China insists on sharing the very authority to define safety discussions [9]. This is a strategy to substitute safety discourse with the issue of global rule-making authority [9]. It is an approach at the level of foreign and security negotiation tactics, rather than a domestic political event in China.
Economic Structure: At the summit, tariff cuts on non-sensitive items worth approximately $30 billion were also finalized [10]. This is a structure where the AI safety dialogue is bundled as a package with a trade truce. The SCMP defined this dialogue body as "the latest formal channel to manage competition" [18]. The analysis suggests that the effect of stabilizing relations is greater than immediate trade achievements [18]. This indicates that AI arms control is not an independent agenda item but a subordinate one dependent on the overall stabilization package of US-China relations.
Security Structure: At the same meeting, expanding communication between the military authorities of both countries was also agreed upon [10]. The AI hotline and the military hotline are designed with virtually similar crisis management logic. This reflects the structural characteristic of combining the military use of AI with cyber threats in emerging and non-traditional security domains. However, this communication channel is limited to mutual notification in the event of accidents [7][9]. It differs in nature from traditional arms control, which restricts capability development itself.
3. Historical Precedents and Comparison of Similar Cases
The comparison target pointed out by the Economic Times is the nuclear Cold War [1]. While competitively building mutually assured destruction capabilities, the United States and the Soviet Union simultaneously established limited crisis management mechanisms, such as hotlines, following the Cuban Missile Crisis. Discussions on the AI hotline explicitly reference this precedent [9]. However, there is a crucial difference between nuclear weapons and AI. Nuclear weapons are possessed by a small number of states and have a verifiable physical reality. AI development is led by private companies, and its proliferation pathways are multi-layered. McGuire's very attempt to apply US-Russia arms control experience to AI [2] carries the challenge of how to overcome these fundamental differences.
A more direct precedent is the US-Russia arms control negotiations during the first Trump administration, in which McGuire himself was involved [2]. Negotiations at the time proceeded by specifying verification systems and mutual notification obligations within the framework of the New Strategic Arms Reduction Treaty (New START). Similarly, the current AI hotline concept started from the minimum requirement of mutual notification in the event of accidents [7][9]. However, unlike New Cold War-style arms control, there is a fundamental difference in the AI domain: the two sides do not share the definition of the term "safety" to begin with [3][9]. While nuclear arms control started from a relatively single axis of destructive power, AI arms control starts with multiple axes—safety, norms, and technological hegemony—intermingled.
Another reference point is the scientific and technological exchange channels during the Cold War. Despite the arms race, the United States and the Soviet Union maintained limited cooperation in space and health sectors. Similarly, the current "Superintelligence Dialogue" between the US and China adopts a method of separating and discussing specific risk areas while maintaining the broader framework of competition [4]. However, an asymmetry surrounding the terminology itself is noticeable. The fact that China has not yet officially used the term "superintelligence" chosen by the Trump administration [4] signals that both sides are attempting to load different agendas into the same dialogue body.
4. Key Variables Shaping Future Developments
The first variable is whether the agenda for the second dialogue meeting, scheduled for November, will be specified [4][7]. While the first agreement was limited to establishing a symbolic framework, whether verifiable notification procedures or classification criteria are presented in the second meeting will serve as a measure of substantive progress.
The second variable is whether industry-led pressure will persist. If additional cases of AI misuse, such as the Australian hacking incident, occur [12], arguments for strengthening regulations could gain traction even within the Trump administration. Conversely, if time passes without further incidents, President Trump's stance of prioritizing competitive advantage is highly likely to solidify.
The third variable is how China utilizes the safety discourse. If China continues to use "safety" as a means to secure global rule-making authority [9], the United States will likely accept multilateral norm discussions only within the framework of countering China. This increases the likelihood that the bilateral dialogue will solidify into parallel conversations with different objectives, rather than developing into a substantive agreement on norms.
The fourth variable is whether semiconductor and AI export controls will continue separately [3][9]. If individual regulations are maintained regardless of symbolic agreements at the leadership level, the AI arms control discourse is highly likely to remain political rhetoric rather than security substance. From the perspective of middle powers, a strategic choice to separately respond to security-linked areas and commercial domains becomes inevitable at the intersection of these four variables.
III. Final Recommended Response Measures
The Rise of US-China AI Arms Control Negotiations: Comprehensive Response Measures
1. Comprehensive Assessment and Recommended Response Measures
Synthesizing the developments since the Washington summit on September 24, US-China AI arms control is highly likely to solidify for the time being into a form where "establishing a dialogue body" and "rejecting substantive regulation" coexist. Although President Trump agreed to establish the superintelligence dialogue [4], he reaffirmed his rejection of joint regulation with China during the meeting with industry executives on September 29 [12][14]. This demonstrates that the White House's cooperative gestures are subordinate to the higher objective of maintaining the domestic industry's competitive advantage. Since China is also adopting a tactical approach of not rejecting safety discussions themselves while demanding the sharing of defining authority as a precondition [9], it must be assessed that both sides prioritize relationship management effects over substantive binding power.
In this structure, the basic stance South Korea should adopt is clear. It must not overinterpret the US-China AI dialogue as the starting point for substantive norm-making. At the same time, South Korea must design its response system under the premise that security-linked technology controls between the two countries will continue regardless of the existence of the dialogue body. This means that South Korea is in a dual position where it is not a direct party in the highest-level issue area of interstate competition surrounding AI, yet must secure a voice in the early stages of norm-making.
The recommended response measures consist of three prongs. First, establish a dual response system that separates security-linked areas from commercial domains. Second, concurrently participate in separate multilateral norm tracks that are not subordinate to the bilateral US-China dialogue. Third, continuously track whether crisis management mechanisms such as the AI hotline and military hotline are actually operational to review the necessity of a South Korean equivalent channel.
2. Short/Medium/Long-Term Action Plans
Short-term (1–3 months): The Ministry of Foreign Affairs and the Ministry of Science and ICT will jointly monitor the outcomes of the second U.S.-China Superintelligence Dialogue meeting scheduled for November [4][7]. How the differences in positions between the two sides regarding the definition of terms (AI vs. superintelligence) are reconciled during the meeting will serve as the primary observation indicator [4]. Concurrently, the Ministry of Trade, Industry and Energy will separately track whether semiconductor and AI export control items are adjusted. Whether the tariff cuts on non-sensitive items worth approximately $30 billion agreed upon immediately after the summit [10] are expanded to AI-related items is a short-term checkpoint for the trade and economic security lines.
Medium-term (6 months–1 year): The Ministry of National Defense and the Ministry of Science and ICT will jointly analyze cases of AI-linked hacking incidents that occurred in places like Australia [12] to pre-design notification and response systems in the event of similar incidents occurring around the Korean Peninsula. They will verify whether the agreement to expand communication between US and Chinese military authorities [10] actually leads to hotline operations, and the Ministry of National Defense will review the necessity of establishing a similar channel between South Korea and the US. Furthermore, they will track whether discussions on applying the US-Russia arms control model to AI, as proposed by McGuire [2], progress to the legislative stage within the United States. This will serve as a leading indicator to capture signs of a shift in the US approach to AI regulation from a bilateral agreement-based model to a domestic legislative model.
Long-term (1-2 years): In preparation for the potential acceleration of discussions on norms for the military use of AI in multilateral channels, including the UN, South Korea should build diplomatic assets that allow it to participate as an independent norm proposer outside the U.S.-China bilateral dynamic. This approach leaves open the possibility that the UN track, currently assessed as having low relevance, could be elevated in the medium to long term. At the same time, South Korea must preemptively establish its own principles and safety standards for the military use of AI to avoid being passively integrated into future standards led by either the United States or China.
3. Monitoring Indicators and Trigger Points
There are four key indicators. First is whether the term 'superintelligence' is officially adopted at the second dialogue meeting in November and China's attitude toward accepting it[4]. If China continues to reject this term, it will signal that the framing competition between the two sides remains unresolved. Second is a shift in the tone of President Trump's public remarks. Whether he shifts his rhetoric from 'maintaining competitive advantage' to the 'need for joint regulation'[14] serves as a leading indicator of substantive policy change. Third is whether there are actual instances of the AI hotline being activated. Whether the incident reporting channel is merely established but remains inactive, or whether both sides respond swiftly in the event of an actual incident, will be a crucial test of the mechanism's effectiveness. Fourth is the trend in changes to the semiconductor and AI export control lists. If controls are maintained or tightened regardless of the dialogue's operation, it would reaffirm that safety dialogues and technological hegemony competition are entirely separate tracks.
The following scenarios can be established as trigger points. If an AI-related security incident occurs in South Korea or an allied nation and an attempt is made to activate the hotline, South Korea should immediately raise the need for its own similar channel in public discourse. Furthermore, if the United States moves to translate AI arms control into domestic legislation, this would signal a shift in the center of gravity from the bilateral negotiation track to the domestic regulatory track; consequently, South Korea's technology cooperation strategy toward the United States must be readjusted accordingly.
4. Summary and Conclusion
Discussions on U.S.-China AI arms control are highly likely to result in a symbolic channel for relationship management rather than a substantive agreement on norms. While maintaining the appearance of cooperation, the Trump administration has not abandoned its competitive advantage frame[14], and China is employing tactics that reframe safety discussions as a matter of rule-making authority[9]. South Korea is not in a position to directly participate in this bilateral dynamic. However, establishing a system to respond separately to security-linked technology controls and commercial AI cooperation, while securing an independent position early on in multilateral norm tracks, is a realistic path that South Korea can take as a middle power. The second dialogue meeting in November and whether the hotline is activated thereafter will serve as the first test to gauge the substance of this trend.
References
[1] [Economic Times] Trump just fired the starting gun on new Cold War with China
[2] [Council on Foreign Relations (CFR)] How to Reach an AI Arms Control Deal With China
[7] [Hürriyet Daily News] US, China agree to open AI incident reporting channel
[8] [Council on Foreign Relations (CFR)] Why AI’s Biggest Rivals Are Suddenly Calling for Restraint
[10] [DW (Deutsche Welle)] China, US agree to set up channel for AI safety incidents
[11] [Council on Foreign Relations (CFR)] The Limits of Trump and Xi’s Photo-Op Summit
[13] [Carnegie Endowment] Who Gets to Put Guardrails on AI?
[15] [Geo News] China, US to open AI 'communication channel' after summit
[16] [The News International] US, China agree on AI ‘communication channel’ after Trump-Xi summit
[17] [Nikkei Asia] Xi's human control or Trump's 'super intelligence': What's next for AI?
[21] [Exame] EUA e China criam canal de comunicação para incidentes com IA
[22] [Business Times (SG)] China media says US shares responsibility for managing AI
[23] [Le Matin du Sahara] Les Etats-Unis et la Chine ouvrent
[25] [The Diplomat] The Trump-Xi Summit Is Over, But Taiwan Is Still Bracing for the Fallout
*This text is an AI translation of an original written in Korean. Some translations or nuances may be inaccurate.
This report is an in-depth analysis planned by an EAI researcher, grounded in sophisticated AI-assisted research, and finalized by the EAI researcher.
How can I help you?