Singapore's Proposal for a UN AI Safety Framework: Middle-Power Norm Competition and Implications for South Korea

Executive Summary

During his speech at the UN General Assembly, Singaporean Foreign Minister Balakrishnan proposed the establishment of a new international organization and a UN framework convention to set safeguards and standards against the backdrop of accelerating US-China AI competition. This is interpreted as strategic positioning by the city-state of Singapore to transform its structural disadvantage in scale competition into norm competition. Balakrishnan's logic—that the UN is the only overarching body with a universal membership structure—stems from the concern that if norm-making is left solely to bilateral negotiations between the US and China, middle powers will be relegated to mere rule-takers. Unless the fundamental distrust between the great powers is resolved, discussions are highly likely to remain at the level of declaring principles, and significant delays are expected before a substantive treaty is concluded. South Korea should pursue early participation in the drafting working group alongside independent domestic AI safety regulatory reforms, while taking into account the constraint that its security alliance structure makes it difficult to directly adopt Singapore-style equidistant diplomacy.

I. Issue Analysis

Singapore's Proposal for a UN AI Safety Framework Convention: Issue Analysis

1. Background and Progress of the Issue

On September 26, Singaporean Foreign Minister Vivian Balakrishnan delivered Singapore's national statement during the General Debate of the UN General Assembly in New York. In his address, he noted the accelerating AI competition between the United States and China, assessing that the stage of calling for a halt in development for the sake of safety has already passed [1]. Instead, he stated the need to establish a new international organization capable of setting safeguards and standards [1]. He remarked, "For new technologies, we must write new rules together, and we must be prepared to create new institutions if necessary" [1]. The concept of a UN-level framework convention was also presented within these remarks [1].

This proposal did not emerge in a vacuum. Throughout this year, Singapore has consistently voiced the perspective of middle powers in AI safety discussions. Prime Minister Lawrence Wong signed a joint statement alongside leaders from more than 20 countries calling for the implementation of frontier AI safety measures [7]. Prior to this, Josephine Teo, Minister for Digital Development and Information, introduced Singapore's approach to AI on a UN platform, emphasizing that the measure of innovation should not be the performance of models but their tangible impact on people's lives [9]. Minister Balakrishnan's latest proposal is an extension of this series of diplomatic positionings.

At the root of this trend lie Singapore's unique structural constraints. As a city-state with a population of 6 million and a land area of 730 square kilometers, Singapore cannot engage in infrastructure and semiconductor competition on the same scale as the US and China [3]. Prime Prime Minister Wong's statement in his National Day Rally speech in August—that Singapore would "harness the power of new technologies, but do it our way"—stems from an acknowledgment of this constraint [3]. The direction of multilaterally designing rules and verification standards aligns with Singapore's strategy to secure a voice in norm competition rather than scale competition.

2. Current Situation

Channel NewsAsia reported Minister Balakrishnan's remarks as a "proposal for a UN framework for global AI safeguards," noting that he listed the categories of risks facing governments, ranging from autonomous systems slipping out of control to the use of AI in developing weapons of mass destruction [4]. He argued that the UN, as the only overarching international organization with a universal membership structure, must play a central role in shaping common rules [4].

Singaporean media does not separate this proposal from the US-China competition. The Business Times reported that during a dialogue with former Australian Prime Minister Kevin Rudd on September 25, Minister Balakrishnan characterized China's AI pursuit as a phenomenon qualitatively different from the Sputnik race of the US-Soviet Cold War [12]. He explained that China is approaching AI with a determination not to repeat its failure to catch up during the Industrial Revolution 150 years ago [12]. This demonstrates that Singapore is positioning itself as an intermediary that explains the logic of both sides rather than aligning with the narrative of either the US or China.

Around the same time, Defense Minister Chan Chun Sing also stated in an interview with local media that "Singapore does not avoid taking sides amidst intensifying US-China competition, but rather makes choices based on its own national interests" [14]. This supports the view that the AI safety framework proposal is not an endorsement of a specific camp, but rather an extension of Singapore's pragmatic diplomacy.

Meanwhile, Washington think tanks are also analyzing the international context in which this proposal was made. The Carnegie Endowment for International Peace pointed out that crisis response frameworks for frontier AI are exposing the limitations of existing approaches [2]. The Brookings Institution addressed the debate over whether the United States can slow down the pace of AI development without falling behind China [10]. The Council on Foreign Relations (CFR) introduced discussions exploring the possibility of a US-China AI arms control agreement [5][8]. In an interview with Foreign Policy, Amandeep Singh Gill, the UN Secretary-General's Envoy on Technology, shared that AI risks and the role of the UN were among the most discussed topics during the high-level week of the UN General Assembly [13].

3. Key Actors and Positions

Singapore's Ministry of Foreign Affairs is the main proponent of this proposal. Minister Balakrishnan defines the UN as a universal multilateral platform, arguing for the necessity of common rules and verification standards within the UN framework, rather than standards set by individual nations or a handful of great powers [1][4]. This is interpreted as an attempt by Singapore—which cannot compete in scale with either the US or China in the semiconductor and data center race—to secure a middle-power stake on the stage of norm design [3].

The United States has not yet announced an official position on this proposal. However, voices within Washington policy circles are increasingly calling for potential AI safety negotiations between the Trump administration and President Xi Jinping [5]. This suggests that concerns over regulatory gaps exist within the US as well. At the same time, given its perception that maintaining a technological edge over China is a matter of national security, the US is expected to take a cautious stance regarding the possibility of multilateral norms constraining its technological superiority [10].

China has independently elevated AI development to a core pillar of its national strategy [6]. The South China Morning Post (SCMP) reported that during a forum held in Singapore, academic and industry figures from both the US and China mentioned the need to establish joint safeguards for AI risk management [11]. However, it remains uncertain whether this consensus will lead to an agreement on a concrete verification framework. China's AI rise is described as a national resolve stemming from the historical trauma of falling behind in the Industrial Revolution [12], foreshadowing that China will not easily yield its sovereign technological autonomy in the process of norm design.

Within the UN Secretariat, the Office of the Envoy on Technology is involved in establishing frameworks to address AI risks. Envoy Amandeep Singh Gill likened the situation to "scorpions in a bottle," highlighting both the limitations and possibilities of the UN's role in a context where mutual deterrence and distrust coexist [13].

4. Key Issues

The first issue is the effectiveness of the proposed new international organization. Although Minister Balakrishnan proposed establishing an organization equipped with verification standards [1][4], it remains uncertain to what extent the US and China will accept external verification of their core technological capabilities.

The second issue is leadership in norm-making. If a UN framework actually reaches the negotiating table, both the US and China are highly likely to try to push through rule designs favorable to their respective interests. The key is whether middle powers, including Singapore, can secure a substantive voice in this process.

The third issue is the coexistence of safety discussions and the logic of competition. Analyses by Brookings and the CFR share the concern that safety measures slowing down AI development could lead to a technological disadvantage relative to competitors [10][5][8]. This suggests the possibility that a framework convention may ultimately remain at a declarative level rather than becoming a binding norm.

The fourth issue is the sustainability of Singapore's positioning. While Defense Minister Chan Chun Sing's remarks on "choices based on national interests" [14] and Minister Balakrishnan's multilateral proposal are complementary, it remains to be seen whether this non-aligned intermediary role can be sustained in the long term under a structure where Beijing's incentives and Washington's pressures operate simultaneously [3].

II. In-Depth Analysis

Singapore's Proposal for a UN AI Safety Framework Convention: In-Depth Analysis

1. Root Cause Analysis

The primary driver of this proposal is the regulatory gap. The starting point is the assessment that the capabilities of frontier AI models are evolving faster than safety verification frameworks [10]. The Brookings Institution frames this gap as a debate between existential risk and immediate risk. While some counterarguments suggest that warnings of catastrophic risks are exaggerated, regulatory lag itself is treated as an undisputed fact [10].

A more direct cause in Minister Balakrishnan's remarks is the perception of timing—specifically, that "it is already too late to call for a halt in development for safety" [1]. This assumes that repeated calls for a development moratorium since 2023 have failed to yield results. Instead, the path he chose is not regulation, but a verification body. The logic that a new international organization is needed to set safeguards and standards stems from the concern that trust cannot be guaranteed solely through individual national self-regulation or voluntary corporate declarations [1][4].

The more fundamental cause is the norm vacuum created by the US-China bipolar structure. While the US and China keep each other in check in the AI development race, they rely on their respective domestic regulatory frameworks for safety issues. A report by the Carnegie Endowment for International Peace summarizes this situation with the question, "Who is putting guardrails on AI?" [2]. Minister Balakrishnan's argument that the UN, as a universal membership body, can fill this vacuum also arises from this context [4]. From Singapore's perspective, there is a calculation that if norm-making is left solely to bilateral negotiations between the US and China, middle powers will inevitably be relegated to rule-takers.

2. Structural Context

2.1 Political Structure: The City-State's Structural Constraints and the Shift to Norm Competition

Behind Singapore's proposal for a UN framework lies the fundamental limitation of its national scale. As a city-state with a population of 6 million and a land area of 730 square kilometers, it cannot match the volume of the data center and semiconductor infrastructure race between the US and China [3]. Prime Minister Lawrence Wong's suggestion in his National Day Rally speech in August of "discipline and execution rather than scale competition" as an alternative stems from an acknowledgment of this constraint [3]. The Business Times evaluated this strategy as "an attempt to change the game to one where Singapore can play a meaningful role and win" [3].

In this context, the proposal for a UN framework can be read as an attempt to substitute volume competition with norm competition. In discussions on verification standards and international organization design, bargaining power and credibility—rather than land area and population—determine one's voice. This is precisely why Singapore has repeatedly raised its voice on multilateral platforms. It is no coincidence that Minister Josephine Teo's UN remarks, Prime Minister Wong's signing of the joint statement on frontier AI safety, and Minister Balakrishnan's UNGA speech all occurred in succession within the same year [7][9][1].

2.2 Security Structure: Avoiding Side-Taking and the Principle of Prioritizing National Interests

On September 22, Defense Minister Chan Chun Sing stated, "Singapore does not avoid taking sides amidst intensifying US-China competition, but rather makes choices based on its own national interests" [14]. This principle is directly reflected in the AI safety framework proposal. Singapore does not fully align with either the US narrative of AI dominance or China's narrative of catching up. In his dialogue with former Australian Prime Minister Kevin Rudd, Minister Balakrishnan explained China's AI pursuit as a "determination not to repeat its failure to catch up during the Industrial Revolution 150 years ago," while distinguishing it from the Sputnik race of the US-Soviet Cold War [12]. This is a rhetorical approach that explains the logic of both the US and China without being confined by the camp mentality of either side.

This security positioning is also intertwined with economic exposure. According to a stress test by the Monetary Authority of Singapore, approximately 32% of Singapore-listed companies would be impacted under a scenario of a sharp drop in AI-related investment [14]. The higher the economic dependence on the AI supply chain, the greater the incentive to reduce international uncertainty surrounding safety norms. The proposal for a UN framework is a policy that emerged at the intersection of security risk management and economic risk management.

2.3 Economic Structure: Preempting Norms and the Hub Strategy

As seen in the case of the AI conference hosted by DBS Bank, Singapore positions itself as a "trusted and neutral intermediary" for AI growth in Asia [15]. This aligns with its financial hub strategy. By participating early in norm-making, Singapore calculates that it can secure a standard-setter status in subsequent discussions on AI governance in the Asian region. The proposal of a universal framework like the UN framework is also closely related to an economic calculation to exert influence over the rule-making process through participation, rather than being subordinate to the standards of a specific camp.

3. Historical Precedents and Comparative Cases

The Treaty on the Non-Proliferation of Nuclear Weapons (NPT) and the experience of establishing the IAEA are cases frequently invoked as reference points for this proposal. In an interview with UN Envoy on Technology Amandeep Singh Gill, Foreign Policy explained the AI safety issue using the Cold War nuclear deterrence metaphor of "scorpions in a bottle" [13]. The argument is that the experience of the United States and the Soviet Union in establishing a verifiable disarmament regime despite the fear of mutually assured destruction could serve as a reference point for designing an AI safety verification framework [13]. However, unlike nuclear technology, AI is developed primarily by private companies and diffuses much faster, making verification significantly more difficult.

The experience of arms control during the US-Soviet Cold War is also discussed. Addressing the possibility of a US-China AI arms control agreement, the Council on Foreign Relations (CFR) cited the perspective of a former official who handled US-Russia disarmament policy at the National Security Council to highlight the difficulties of designing verification mechanisms [5]. Minister Balakrishnan himself distinguished China's AI pursuit from the Sputnik race [12]. This can be read as an attempt to acknowledge Cold War similarities while emphasizing the distinct nature of today's AI competition—namely, that it is a diffuse competition involving multiple actors rather than a simple bipolar structure.

The international climate change regime is another point of comparison. The UN Framework Convention on Climate Change (UNFCCC) assumed universal participation, but establishing a binding verification framework took decades. If the AI safety framework follows this path, it is highly likely to remain at the level of declaring principles initially, with verification and implementation mechanisms deferred to subsequent negotiations. The fact that Singapore is already experiencing tension between expanding its domestic AI data centers and its climate policy supports this comparison. The pattern where the expansion of AI infrastructure increases energy demand and collides with the climate diplomacy agenda is a structural pattern also observed in cases of backlash against data centers in the United States.

4. Key Variables Shaping Future Developments

The first variable is the response of the United States and China. For a UN framework to have substantive binding force, at least both the US and China must demonstrate a willingness to participate. The CFR reports that voices calling for a US-China AI arms control agreement are growing within the US [5]. However, this is a matter for bilateral channels between Washington and Beijing, and does not necessarily require going through a multilateral framework like the UN. If the US and China establish a verification framework through separate bilateral channels without utilizing the UN framework, Singapore's proposal risks being relegated to symbolic agenda-setting.

The second variable is the design of the new international organization. Minister Balakrishnan only mentioned that "we must be prepared to create new institutions" without presenting a concrete organizational structure or scope of authority [1]. Its effectiveness will vary significantly depending on whether it aims for an organization with IAEA-style inspection powers or remains at the level of a consultative body limited to information sharing and standard recommendations. How to justify the creation of a separate organization when the UN Office of the Envoy on Technology already exists is also a key question [13].

The third variable is the mode of private sector participation. AI development is driven not by states, but by a handful of big tech companies. Although there was an unprecedented instance of heads of leading AI labs jointly calling for safety restraint, it remains uncertain whether this will translate into intergovernmental agreements [8]. The joint statement by leaders of more than 20 countries, including Prime Minister Lawrence Wong, was also formatted to urge implementation by both companies and member states [7]. The extent to which an intergovernmental agreement can secure voluntary compliance from corporations will determine the substantive weight of the framework.

The fourth variable is alignment from ASEAN and other middle powers. A proposal by Singapore alone has limitations in becoming an official agenda item at the UN General Assembly. Whether ASEAN member states or European nations co-sign this proposal or develop it into a separate resolution will determine the issue's momentum. In the data reported so far, individual remarks by Singaporean government officials remain central, and the formation of a multilateral support base is still in its early stages.