UN Security Council AI Safety Briefing and US-China Governance Competition: China's Dual-Track Response and Implications for Middle Powers

Executive Summary

In September 2026, the AI deceleration discourse initiated by Anthropic spread as an international governance agenda through a UN Security Council briefing and a US-China summit. However, the consensus on frontier AI cooperation identified at the Security Council stalled at the implementation stage, with the United States maintaining a firm stance against international regulation itself. While China reframes the US-led deceleration argument as a struggle for hegemony, it simultaneously pursues a dual-track approach, with its Ministry of State Security designating US AI models as national security threats. As safety discourse is highly likely to continue being consumed as a bargaining chip rather than facilitating substantive cooperation, South Korea needs to secure separate communication channels to avoid being excluded from US-China bilateral tracks and manage its public and private sectors separately.

I. Issue Analysis

China's Position Amid the UN Security Council AI Safety Briefing and the Global Governance Debate

1. Background and Progress

The flashpoint of this issue was a post by Anthropic CEO Dario Amodei on September 12, 2026 [9]. It was titled "We Need to Pace Frontier Development" [9]. Amodei warned that autonomous AI agents could take over the internet via botnets within 6 to 12 months [9][14]. With OpenAI's Sam Altman and SpaceX's Elon Musk immediately expressing agreement, this statement quickly spread beyond Silicon Valley to become an international political agenda [13][10].

Against this backdrop, a UN Security Council meeting, which had been prepared under French leadership since August 2026, was held on September 23, 2026 [15]. Under the agenda of "Maintenance of International Peace and Security," it was recorded as an unprecedented scene where heads of AI companies, including Altman, personally attended and briefed the Security Council [14]. Le Monde reported that during this meeting, the private sector and the vast majority of countries reached a rare consensus on the need for international cooperation on frontier AI models [15]. However, it pointed out that this consensus immediately stalled at the implementation stage because the United States maintained a firm stance against any form of international regulation [15].

This meeting did not take place in a vacuum. It was held just two days before the US-China summit scheduled for September 24, 2026, in Washington [10]. On September 14, 2026, two days after Amodei's call for deceleration, President Trump labeled AI doomsday scenarios as a "scam" and remarked that amid the backlash against AI and data centers, "only China is happy" [10]. This remark served as the backdrop for the US-China summit throughout President Xi Jinping's visit to Washington [10][7]. The Asahi Shimbun reported that as Amodei's plea to "slow down the pace of improving AI model capabilities" garnered global attention, the focus of the summit shifted to how far the two AI superpowers, the US and China, could compromise to ensure safety [7].

2. Current Situation

Beijing's external response was swift and consistent. Foreign Ministry Spokesperson Guo Jiakun refuted Amodei's deceleration proposal immediately after it was made [3]. China believes that the US-led deceleration argument is predicated on commercial interests and maintaining technological superiority over China [9][17]. The News International recently reported that while Beijing publicly rejected the Anthropic CEO's proposal for "global frontier pacing," it quietly updated and released its own AI risk framework at the same time [1]. This represents a dual-track response: externally refuting the US-led deceleration argument as a framework of hegemonic competition, while internally establishing its own risk management system.

This dual structure is even more pronounced in the remarks of Minister of State Security Chen Yixin. The head of this agency, which serves as both an external intelligence service and an internal dissident surveillance apparatus, specifically targeting OpenAI and Anthropic models as political, ideological, and cyber security threats is evaluated as "the most concrete manifestation of security perception Beijing has released to date" [3]. In other words, while external rhetoric refutes the US demand for deceleration as a hegemonic competition frame, a separate internal track is simultaneously underway, defining US AI models themselves as security threats [3].

On the US side, President Trump flatly rejected the concept of international AI regulation in his UN General Assembly speech, characterizing it as a "globalist project to control AI" [12]. At the same time, he rebranded the technology as "superintelligence" in official US terminology and expressed a strong commitment to securing a competitive edge, using the phrase "the winner takes the spoils" [12]. The Economic Times expressed concern that this trend could lead to a collusion between the US and China alone, resulting in a structure where other countries' voices are excluded [12].

Despite this confrontation, working-level channels are moving separately. According to Wired, US and Chinese officials have begun discussing a mechanism to mutually notify each other of national security threats posed by AI, the so-called "US-China AI Dialogue" [17]. This demonstrates a structure where rhetorical confrontation at the leadership level over the language of safety runs parallel to the establishment of crisis management channels at the working level [6][17].

3. Key Actors and Positions

Chinaexpresses agreement with the UN Secretary-General's call for international cooperation, but opposes having the framework of that cooperation follow terms set by the United States. Beijing's logic advocates for "balanced global governance," arguing that the authority to define safety must be shared rather than monopolized by the US [6]. Commentaries from state-run media present the condition that the same standards must apply to US companies as a prerequisite [6]. This is a strategy to effectively substitute discussions on technological risk with the issue of global rule-making authority [6]. At the same time, the Ministry of State Security's threat perception remains separate from external rhetoric, maintaining an internal track that treats US AI models as specific political and cyber security threats [3].

United Statessees divided positions between the administration and the industry. The Trump administration rejects international regulatory discussions out of concern that they could undermine its competitive advantage, defining AI as an asset in its hegemonic competition with China [12][10]. Conversely, industry leaders such as Altman, Amodei, and Musk personally participated in the UN Security Council briefing to appeal for the necessity of international cooperation [14][15]. A structure is also observed where the White House's priority on competition with China operates separately from the pressure tracks of regulatory agencies like the FCC and Congress [6].

UNis the actor that provided the unprecedented stage of the Security Council. Prepared since August under French leadership, this briefing is recorded as the first instance where the Security Council addressed AI under the agenda of "Maintenance of International Peace and Security" [14][15]. The Secretary-General's call for international cooperation superficially resonates with China's discourse on balanced governance, but due to US opposition, no concrete implementation plans were produced [15].

4. Key Issues

The first issue is the authority to define the concept of "safety." While both the US and China refer to safety, they define it differently. The US defines the potential misuse of AI as a tool for censorship and oppression as a safety threat. China defines safety as preventing AI from becoming a "runaway horse"—that is, preventing the technology itself from running wild—and establishing balanced international governance [6]. The second issue is the leadership over the international regulatory framework. The consensus formed in the multilateral arena of the UN Security Council has failed to translate into the implementation phase due to its clash with the US bilateral priority on competition with China [15][12]. The third issue is the gap between discourse and implementation. While rhetorical confrontation continues at the leadership level, working-level officials from both countries are exploring separate crisis management channels, such as a mutual notification mechanism for AI threats [17][9]. These three strands of tension are expected to diverge into paths of either deepening governance fragmentation or limited institutionalization, depending on the follow-up measures of the US-China summit and the implementation of Security Council discussions.

II. In-Depth Analysis

China's Position Amid the UN Security Council AI Safety Briefing and the Global Governance Debate

1. Root Cause Analysis

The root cause of this debate lies in the fact that the AI safety discourse did not start as a pure technological risk management agenda from the beginning. Looking at the developments following Amodei's call for deceleration, the language of safety was consumed by both the US and China as a bargaining chip and a means to secure a competitive advantage [9]. On September 14, 2026, two days after Amodei's remarks, President Trump labeled AI doomsday scenarios as a "scam" [10]. At the same time, amid the backlash against AI and data centers, he remarked that "only China is happy" [10]. The very positioning of this remark demonstrates that the safety debate within the US remains locked within the framework of maintaining a competitive edge over China.

China's response logic shares the same structure. Foreign Ministry Spokesperson Guo Jiakun immediately refuted Amodei's deceleration proposal [3][9]. Beijing believes that the US-led deceleration argument is predicated on maintaining US companies' superiority over China [9][17]. The CFR points out that behind this backlash was an unusual consensus forming among US AI labs [13]. In other words, a structure of mutual misperception forms the starting point of the debate, where US companies' calls for self-regulation are read by China as an attempt to hold back competitors, and China's refutation is in turn read by the US as a challenge to its hegemony.

The root cause becomes even more complex when overlaid with the stage of the UN Security Council. Le Monde reports that a consensus on the need for frontier AI cooperation was formed between the private sector and the vast majority of countries during the Security Council meeting [15]. However, this consensus stalled the moment it transitioned to implementation discussions [15]. This was because the United States maintained a firm stance against any form of international regulation [15]. Ultimately, the root cause is not a difference in risk perception, but a sovereign struggle over who holds the authority to define and regulate risk.

2. Structural Context

Political Structure: The Dual-Tracking of Safety Discourse

China's response exhibits a dual structure where external rhetoric and internal threat perception operate separately. Minister of State Security Chen Yixin specifically identified OpenAI and Anthropic models as political, ideological, and cyber security threats ahead of the summit [3]. The head of this agency, which functions as both an external intelligence service and an internal dissident surveillance apparatus, making such remarks is evaluated as "the most concrete manifestation of security perception Beijing has released to date" [3]. At the same time, China maintains its external rhetoric refuting the US-led deceleration argument as a framework of hegemonic competition [1][9]. The News International reported that while Beijing publicly rejected Amodei's proposal, it quietly updated its own AI risk framework [1]. This is a structure that speaks of cooperation and balance externally, while internally guarding against US models as tools for censorship and ideological infiltration. This explains why China, while expressing agreement with the Secretary-General's call for international cooperation at the UN, seeks to cooperate on regulation only to the extent that it does not infringe upon its own sovereignty and censorship system.

The structure on the US side is also not monolithic. The White House's summit cooperation track and the pressure tracks of regulatory agencies like the FCC and Congress are already operating separately [6]. In his UN General Assembly speech, President Trump rejected international regulation, defining it as a "globalist scheme to control AI" [12]. At the same time, he rebranded AI as "superintelligence," appropriating it for official US purposes [12]. This structure explains the duality where the US agrees to organize briefings by AI company heads at the UN Security Council while opposing the formation of binding international norms.

Economic and Technological Structure: Frontier Competition and the Asymmetry of Deceleration Arguments

In the economic structure, the key point is that the competitive relationship among US AI companies undermines the credibility of their calls for deceleration. The SCMP points out that while Amodei, Altman, and Musk speak of deceleration, they remain locked in high-intensity competition with each other and with China [16]. In other words, there is no structural incentive for calls for deceleration to translate into an actual slowdown in development. Brookings raises concerns at this juncture that fear of existential risk could overshadow immediate threats [2]. This asymmetry opens up a favorable interpretive space for China, as it provides grounds to define the US calls for deceleration not as substantive safety measures, but as attempts to secure a rhetorical advantage.

Security Structure: Involvement of National Security Agencies and the Dual-Use Technology Issue

Another structural feature is that AI has already been incorporated into the official agenda of national security agencies. According to Wired, US and Chinese officials have begun discussions on a mechanism to mutually notify each other of AI-related national security threats, the so-called US-China AI dialogue system [17]. While this concept is similar to Cold War-era hotlines or accident notification systems, it faces a fundamental difficulty: due to the dual-use nature of AI, the boundary between military threats and commercial civilian services is blurred. The very fact that the Ministry of State Security's threat perception targets commercial chatbot models suggests that the establishment of a security dialogue system could, in practice, expand into the realms of censorship and information control [3].

3. Historical Precedents and Comparative Case Analysis

The structure of this debate partially overlaps with patterns from the formative period of the nuclear arms control regime. Maguire of the CFR, discussing ways to reach a US-China AI arms control agreement, mentions his past experience as the State Department's working-level head of US-Russia arms control policy [5]. This shows that attempts to approach AI safety issues through the framework of Cold War strategic arms negotiations are already underway in Washington policy circles. However, unlike nuclear weapons, AI is decisively different in that its development is driven by private companies rather than states. The very fact that those who personally briefed the Security Council were Altman and Amodei, rather than state representatives [14][15], exposes the limitations of the current system, which has no choice but to rely on corporate self-regulation pledges rather than verifiable interstate agreements.

The similarity to the structure of international climate change negotiations is also worth noting. A separate CFR analysis evaluates that the AI safety crisis creates a sense of dj vu with the early phases of globalization [8]. In other words, it begins with concerns raised by a small number of leading countries and companies, gradually spreads to discussions in international organizations, but binding agreements are ultimately delayed due to concerns over national industrial competitiveness. The phenomenon where the consensus on cooperation stalled at the implementation stage during the UN Security Council briefing [15] mirrors the recurring gap between declaratory agreements and substantive reduction targets in climate change negotiations.

However, there is a distinct difference in this case. The fact that France proactively prepared this Security Council meeting since August 2026 [15] demonstrates an attempt by a third-party actor to intervene and avoid being trapped in a US-China bilateral framework. The Economic Times points out the risk that, ahead of the Trump-Xi summit, US-China bilateral negotiations could culminate in a "deal between two superpowers" that excludes the voices of other nations [12]. This suggests that the historical pattern of Cold War US-Soviet arms control, which proceeded primarily through bilateral treaties while excluding the opinions of third-world countries, is showing signs of repeating in AI governance.

4. Key Variables Shaping Future Developments

The first variable is whether the US-China AI dialogue system becomes institutionalized. The key is whether the discussions on a mutual notification mechanism for national security threats reported by Wired [17] actually develop into a regular channel. Whether this channel is confined to security accident notifications or expands to commercial model regulatory standards will shape the nature of governance.

The second variable is whether China's dual-track response persists. If the structure of advocating for balanced international cooperation externally while defining US models as security threats internally [3][1] is maintained, it will be difficult for the UN-level consensus on cooperation to lead to substantive regulatory agreements. Conversely, if the Ministry of State Security's threat perception softens or the level of public disclosure decreases, this could be interpreted as a signal for expanding cooperation.

The third variable is the alignment between policy tracks within the United States. As long as the current structure remains—where the White House's cooperative rhetoric operates separately from the pressure stance of Congress and regulatory agencies [6]—symbolic agreements reached at summits or Security Council briefings are highly likely to continue independently of individual regulatory measures such as semiconductor export controls. Whether this gap narrows is key to substantive implementation.

The fourth variable is whether the involvement of third-party state groups expands. Whether the French-led Security Council briefing [15] remains a one-off event or serves as a catalyst for the EU and other middle powers to continuously engage in the norm-shaping process will determine the future governance landscape. If it culminates in a US-China bilateral deal [12], the space for middle powers, including South Korea, to participate in norm-shaping will shrink accordingly.