Escalating US-China Tensions over AI Competition and Safety Dialogue: The Pre-Summit Contest over Norms and South Korea's Response

Executive Summary

Ahead of the September 24 US-China summit in Washington, the two countries have discussed establishing an AI hotline. However, a substantive agreement on norms is unlikely, as each side imbues the term 'safety' with different political objectives. President Trump has responded to industry warnings about AI risks not by strengthening regulations but by creating a new security organization, the 'AI Force.' China has rejected US-led calls to slow down development while simultaneously fostering greater risk awareness internally. Due to a bifurcated US policy structure—where the White House's cooperation track operates separately from the pressure track of regulatory agencies and Congress—individual regulations like semiconductor and AI export controls are likely to persist regardless of any symbolic summit agreements. Coupled with the structural red line of the Taiwan issue, the AI safety dialogue will likely be treated as a secondary outcome of the summit. The South Korean government and companies should establish a dual-track response system that distinguishes between security-related sectors and general commercial areas. Rather than hastily aligning with US AI standards, they need to build their own capabilities while concurrently reviewing the stability of their contracts.

I. Analysis of the Current Situation

Escalating US-China Tensions over AI Competition and Safety Dialogue: Analysis of the Current Situation

1. Background and Developments

The upcoming September 24 Trump-Xi Jinping summit in Washington is a follow-up to agreements made during Trump's visit to Beijing in May [9]. President Trump's personal announcement that a White House state dinner is being prepared indicates that this summit differs in nature from the 2023 APEC summit in San Francisco [9][3]. While the latter was aimed at mending relations strained by the spy balloon incident, this meeting builds on the May visit to emphasize continuity in managing the relationship [9].

An intergovernmental dialogue channel on AI is not in itself new. The two leaders had already agreed to launch such talks at their May summit in Beijing [3]. Ahead of the Washington summit, on September 20, delegations led by US Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng met at the JPMorgan Chase headquarters in New York to discuss an AI-related communication mechanism [1][17]. The US proposed establishing a so-called "AI hotline" for mutual notification in case of an AI incident serious enough to threaten national security [17]. Jamieson Greer from the Office of the US Trade Representative was also present at the meeting [1].

Underlying this trend is a sense of crisis originating from the industry itself. Warnings that AI could pose an existential threat to humanity gained international attention after figures like Anthropic CEO Dario Amodei and other researchers repeatedly argued for slowing the development of frontier AI, a position they call "pace the frontier" [2][7][16]. In the run-up to the US-China summit, these warnings have brought the two countries' AI competition into sharp focus [7].

2. Current Situation

China has effectively rejected US-led calls to slow the pace of AI development. The *Nihon Keizai Shimbun* (Nikkei) reported that China is pushing back against these demands, conveying expert views that mistrust between the two countries will fundamentally limit the scope of cooperation on safety [7]. The Council on Foreign Relations (CFR) analyzes that concern over AI risks is also growing within Beijing, citing an unusually strong warning from the head of China's Ministry of State Security that AI could threaten the Communist Party's rule [8]. However, the general consensus among experts on the ground is that it remains uncertain whether these internal concerns will lead Beijing to accept the US-proposed framework for slowing down [8].

On September 19, President Trump announced a plan to create an 'AI Force' via his social media platform, Truth Social [6][19]. The announcement of a new security organization through the president's personal social media, rather than official Pentagon or White House channels, was itself unusual [6]. No specific details were provided on its organizational structure, departmental affiliation, or legal authority [6][4]. Trump added that he would soon appoint an 'AI Czar' to lead the body and compared its name to the Space Force, a signature achievement of his first administration [6][4]. EAI analysis assesses this move as "a political response strongly aimed at neutralizing industry warnings about AI risks by creating a security organization instead of imposing regulations" [6].

This announcement is putting pressure on the safety dialogue itself. The Dutch newspaper *NRC* reported that Trump's AI Force concept is straining the upcoming safety summit with China and that the announcement, despite its lack of specifics, is already generating concern and backlash in China [15]. The *South China Morning Post* (SCMP) cited experts who suggested that while the AI Force could provide Beijing with a clear counterpart for the safety dialogue, it also risks becoming a tool to tighten the technology blockade and export controls against China [4].

*The Hankyoreh* reported that AI and tariffs will be key items on the summit agenda, with discussions planned on trade, tariffs, rare earth supply chains, and artificial intelligence [14]. The expected attendance of numerous Big Tech CEOs at the state dinner further indicates that the summit is centered on a technology agenda [14][18].

3. Key Actors and Positions

United States (Trump Administration)The administration frames the safety discussion solely within the context of maintaining a competitive edge over China [3]. On September 13, Trump declared, "whoever wins AI, wins" [18]. While directly rejecting industry calls to slow down, the administration has separately proposed a mechanism for mutual notification of AI incidents that pose a national security threat—the hotline [17][1]. This can be interpreted as an attempt to co-opt the concept of safety into a security and military framework rather than a regulatory one [6]. With the White House's cooperation track already operating separately from the pressure track of regulatory agencies like the FCC and Congress, individual export controls are likely to continue to be implemented, irrespective of the summit's outcome [3][9].

ChinaThrough commentaries in state-affiliated media, China has set preconditions for the dialogue: shared authority in defining 'safety' and the application of the same standards to US companies [3]. This represents a strategy to reframe the discussion on technological risk as a contest over global rule-making authority [3]. While internal concerns about AI risks exist, as evidenced by the warning from the head of the Ministry of State Security [8], Beijing externally maintains its refusal to accept the US-led push to slow down development [7]. Having already established a red line—that it could cancel the summit if new arms sales to Taiwan are approved [9]—China is highly unlikely to be drawn into an American-defined framework for the AI safety dialogue.

US AI Industry(e.g., Anthropic) is voicing a different perspective from that of the government. The Carnegie Endowment for International Peace highlights the dual position of companies like Anthropic, which raise safety concerns while also providing substantive support for the US government's integration of advanced AI systems [11]. In essence, the industry's call to slow down and the government's securitizing response are proceeding on separate tracks.

South KoreaWhile not a direct party to this dynamic, EAI analysis suggests that the South Korean government and companies should "concurrently build up their own capabilities and review the stability of their contracts" rather than hastily aligning with US standards [6]. This calls for a dual-track approach that separates risk management in security-linked sectors from the maintenance of existing cooperation in the commercial domain [3][9].

4. Key Issues

The first issue is the authority to define the concept of 'safety.' The United States seeks to narrowly frame safety at the level of a notification mechanism for national security threats, while China aims to expand the definition of safety itself into an issue of shared global norms [3][17]. As these two approaches target fundamentally different goals, the concept is likely to be used as a bargaining chip by each side rather than leading to a substantive agreement [3].

The second issue is defining the nature of the AI Force. Opinions among experts are divided on whether it will become a cooperative partner in the safety dialogue or a new tool for the technology blockade against China [4][15]. Since it was announced without a clear legal basis or organizational substance, future follow-up actions will be key to determining its true character [6].

The third issue is the separation of the summit diplomacy track from the regulatory track. Even if a symbolic agreement is reached at the summit, individual regulations such as semiconductor and AI export controls are likely to continue independently of the White House track [3][9]. Coupled with the situation where the Taiwan issue has solidified as a structural red line, the AI safety dialogue is highly likely to be treated as a secondary outcome of the summit [9][3].

II. In-Depth Issue Analysis

Escalating US-China Tensions over AI Competition and Safety Dialogue: In-Depth Issue Analysis

1. Analysis of Root Causes

At the root of this war of nerves lies a fundamental disagreement over the concept of 'safety' itself. EAI analysis notes that "it is difficult to reach a substantive agreement on norms because the two countries imbue the term 'safety' with different political objectives" [3]. For the Trump administration, safety is little more than rhetoric for maintaining a competitive edge over China, evidenced by its outright rejection of industry calls to slow down development [3]. Beijing, in contrast, seeks to reframe the safety discussion as a matter of global rule-making authority. EAI analyzes that China has "presented the sharing of authority to define safety and the application of the same standards to US companies as preconditions" [3]. In effect, they are using the same word with diametrically opposed negotiating goals.

This gap is not accidental but stems from the domestic political structures of both countries. In the United States, industry figures like Anthropic CEO Dario Amodei have warned of AI's existential risks, calling to "pace the frontier" [2][16]. President Trump, however, did not take these warnings as a basis for strengthening regulation. Instead, he responded with a political maneuver: the creation of a new security organization. EAI defines the 'AI Force' announcement as "a political response aimed at neutralizing industry-led warnings about AI risks by creating a security organization rather than imposing regulations" [6]. This is an unusual combination of deregulation and securitization occurring simultaneously.

The dynamics in China are different. The CFR sees risk awareness spreading within Beijing, citing the fact that the head of China's Ministry of State Security warned that AI technology could threaten the Communist Party's rule itself [8]. Nevertheless, China is not accepting the slowdown framework proposed by the US. Nikkei explains this as a structure where deep-seated mistrust between the two countries limits the scope of cooperation [7]. From China's perspective, accepting US-led calls to slow down could be seen as submitting to American norms while a technology gap persists. *Foreign Policy* diagnoses that there is suspicion within Beijing that the demand for a slowdown, made in the name of safety, is actually an attempt to lock in a US competitive advantage [10].

2. Structural Context

Political Structure: This war of nerves is unfolding within a structure where the summit diplomacy track and the regulatory agency track operate separately. EAI analyzes that "with the White House's cooperation track already operating separately from the pressure track of regulatory agencies like the FCC and Congress, individual regulations such as semiconductor and AI export controls are likely to continue, even if a symbolic agreement is reached at the summit" [3]. Because of this dual-track structure, friendly gestures at the summit do not guarantee substantive policy changes. Trump's proposal at the New York meeting for an 'AI hotline'—a mutual notification system for AI incidents that rise to the level of a national security threat [17]—also appears to be a separate track that, despite its cooperative veneer, proceeds independently of individual regulations like export controls.

Layered on top of this is the structural red line of the Taiwan issue. EAI believes that "with the Taiwan issue solidified as a structural red line, the AI safety dialogue is likely to be treated as a secondary outcome of the summit" [3][9]. Since Beijing has taken the position that it could cancel the summit if new arms sales to Taiwan are approved [9], the AI safety dialogue is inherently at risk of being relegated to a secondary agenda item.

Economic Structure: The AI competition is now brought to the negotiating table bundled with issues of tariffs and rare earth supply chains. According to *The Hankyoreh*, the Washington summit is scheduled to include discussions on trade, tariffs, rare earth supply chains, and artificial intelligence, with numerous US Big Tech CEOs attending the state dinner [14]. This shows that AI is not being treated as a pure tech-security issue, but as part of a negotiating package entangled with trade and industrial policy. Trump's declaration on September 13 that "whoever wins AI, wins" [18] also frames AI as the front line of the competition for economic hegemony.

Security Structure: The 'AI Force' concept is more about symbolic politics than a substantive security organization. The SCMP, citing experts, reports that while this organization could become a clear counterpart for the AI safety dialogue with China, it also risks being turned into a tool to further tighten the technology blockade and export controls [4]. The Dutch newspaper *NRC* also reported that Trump's 'AI Force' plan is acting as a pressure point on the safety summit with China, and that this rhetoric is provoking concern and backlash within China [15]. The very method of eliciting a reaction from the other country with just a name and an announcement format, while the organization's substance remains unclear, shows that this issue is closer to political signaling than to substantive arms control.

3. Historical Precedents and Comparative Cases

President Trump's direct comparison of the 'AI Force' to the creation of the Space Force, a signature achievement of his first administration, is more than a rhetorical choice [6][4]. The Space Force set a precedent by organizing an emerging technological domain into an independent body separate from the existing military structure, thereby maximizing its political symbolism. The 'AI Force' concept can similarly be read as a project intended to repeat a narrative of the president's personal security achievements, rather than focusing on organizational substance [6]. However, whereas the Space Force was established with a legal basis through congressional approval and the National Defense Authorization Act process, the 'AI Force' began with an announcement on the president's personal social media, leaving its departmental affiliation and legal authority unclear [6][4]. This procedural difference raises questions about the future effectiveness and sustainability of the organization.

A comparison with the US-Soviet nuclear arms control negotiations during the Cold War is also instructive. At that time, the US and the Soviet Union were able to establish hotlines and verification systems based on the shared fear of mutually assured destruction. In the current US-China AI competition, however, a comparable level of shared threat perception has not been formed. *Mint* points out the coexistence of conflicting motives: the US AI industry wants to slow down, while President Trump wants to accelerate to maintain an edge over China [12]. In the Cold War nuclear negotiations, there was at least a consensus between the governments on the need for de-escalation and control. Today, this consensus has not even been formed within the US government between the industry and the White House. In this respect, the current situation more closely resembles a structure where norm formation is slow, rather than a traditional arms control precedent.

4. Key Variables Shaping Future Developments

The first variable is whether the 'AI Force' is legally institutionalized. Whether it becomes a substantive tool for military and security integration or remains mere symbolic rhetoric will depend on whether its departmental affiliation, budget, and command structure are specified [6][4].

The second variable is whether the separation between the summit track and the regulatory agency track continues. EAI believes it is highly likely that individual export control measures will continue to accumulate separately, regardless of any symbolic agreements from the summit diplomacy [3][9]. As long as this bifurcated structure persists, it will be difficult to judge substantive policy changes based on the summit's announced outcomes alone.

The third variable is a potential escalation of the Taiwan issue. With a red line established that Beijing could cancel the summit if new arms sales to Taiwan are approved [9], an escalation of this issue could push the AI safety dialogue off the negotiating table entirely.

The fourth variable is the pace at which AI risk awareness spreads within China. Beijing's negotiating posture could change depending on whether the warning from the head of the Ministry of State Security leads to a policy shift or remains rhetoric for external negotiations [8].