[International Politics in the Age of AI] ⑫ AI Competition and the Crisis of International Norms? Dynamics and Challenges of Governance Competition for Lethal Autonomous Weapons Systems

I. The Advent of AI Warfare and Norms for Lethal Autonomous Weapons Systems

While there are differing views on whether the impact of AI on military innovation will be radical or incremental, there is little disagreement that the military use of AI will be a decisive factor on future battlefields.[1] The military use of AI is already a reality. In January 2026, it appears that Palantir used Anthropic's AI model, Claude, during a special operation to apprehend Venezuelan President Maduro.[2] The Russia-Ukraine War, and the Iran War which began on February 28, 2026, with attacks by the United States and Israel on Iran, demonstrate the full-scale militarization of AI across air defense systems, drones, and command and control systems.[3]

As AI is increasingly used on the battlefield, competition to enhance the capabilities of military AI, alongside conflicts over values and norms regarding the legitimate military use of AI, are becoming pressing issues. In February 2026, tensions escalated between the U.S. Department of Defense and Anthropic over the military use of AI. Anthropic was the only AI model company approved for use in military secret systems and had signed a contract worth $200 million with the U.S. Department of Defense in July 2025. However, after the operation to apprehend Venezuelan President Maduro, tensions flared when Anthropic inquired with Palantir about the use of Claude during the operation.[4]

This was because Anthropic adhered to the principle that the use of its AI model, Claude, was unacceptable for large-scale surveillance of U.S. citizens and for use in lethal weapons without human intervention, which conflicted with the Department of Defense's direction for aggressive military utilization of AI. On February 27, just before the U.S. attack on Iran, President Trump posted on his social media that Claude would be removed from all federal agencies, and[5], following a warning from Secretary of Defense Pete Hegseth on February 24, the Department of Defense officially designated Anthropic as a supply chain risk company on March 4.[6] The Department of Defense also maintained that Anthropic's ethical constraints were not only "excessively restrictive" but also "fundamentally impossible to implement considering the legal complexities of military operations and the realities of the battlefield."[7]

The conflict between the U.S. Department of Defense and Anthropic regarding the military use of AI cannot be viewed solely as a contractual issue between the DoD and an AI company. Narrowly interpreted, it is an event that could weaken the position and influence of companies favoring the ethical development and use of AI in terms of competition among AI models. Other companies within and outside the AI ecosystem seeking contracts with the U.S. government, such as Amazon Web Services and Nvidia, as well as governments and companies of other countries seeking contracts with these companies, would need to reconsider their contracts with Anthropic.

However, a more significant issue is that the conflict between the U.S. Department of Defense and Anthropic represents a kind of "stress test" for the direction of military AI development, which could determine the future landscape of AI-related battlefields.[8] It is also a process of establishing a compass for the fundamental goal: navigating the path between "what can be done" and "what should be done" in the development and use of military AI, the logic of consequence versus the logic of appropriateness, and the amoral development for competitive victory versus ethical and responsible development and use. In particular, the fact that one of the areas Anthropic refuses to permit AI use is fully autonomous weapons indicates that the conflict between the U.S. Department of Defense and Anthropic is closely linked to the governance competition surrounding Lethal Autonomous Weapons Systems (LAWS), which the international community is currently discussing for legalization.[9]

II. AI LAWS Governance Competition: Principles and Gaps

Global AI governance competition has been characterized by competition between the United States, which favors a business-friendly approach, China, which favors a government-centric approach, and the EU, which advocates for a rights-based approach.[10] In 2023, the EU published the EU AI Act, which categorizes regulations and prohibitions based on the potential risks posed by AI use. The Biden administration in the U.S. also published the "Blueprint for an AI Bill of Rights" in October 2022 and the "Executive Order on Safe, Secure, and Trustworthy Development and Use of AI" in October 2023, reflecting a significant portion of the rights-based approach advocated by the EU. However, with the advent of the second Trump administration, the direction shifted towards deregulation and fostering a pro-innovation environment for businesses, with the revocation of Executive Order 14110 on the first day of the term and the announcement of the Stargate Project. Concurrently, due to the impact of China's DeepSeek-R1 launch in January 2025, even the EU, which had advocated for a rights-based approach in digital norm discussions, is now pursuing AI capability development and technological innovation. On November 20, 2025, the "Digital Simplification Package" was announced, which postponed the application of the EU AI Act, originally scheduled for August 2026, to December 2027, a delay of 16 months, and relaxed personal data protection regulations. Ultimately, while AI governance competition seemed to be converging towards the EU's rights-based approach based on the Brussels effect, recently, the competition for norms between the U.S. and China, focused on gaining a first-mover advantage in AI competition rather than ethical and responsible AI development and use, is intensifying.

However, the dynamics of norm competition in the governance of LAWS, one of the most contentious and urgent areas of norm competition surrounding AI, show significant differences from the general AI governance competition.[11] Unlike the AI governance competition, where clear differences exist among major actors such as the U.S., China, and the EU, in the LAWS governance competition, major technology-leading states like the U.S. and China are not advocating for distinct norms within the spectrum from complete prohibition and regulation to complete allowance of LAWS. Instead, they are showing implicit cooperation to ensure that the ongoing LAWS governance concludes in a competition-friendly manner.

Unlike the fragmented discussions on AI governance occurring in various international organizations and multilateral forums, the current focus of LAWS governance discussions is within the Group of Governmental Experts (GGE) on LAWS at the UN Convention on Certain Conventional Weapons (CCW). Major countries, including the U.S. and China, have stated that the UN CCW GGE is the most appropriate forum for LAWS governance discussions. For example, in September 2025, the U.S., along with key allies and partners including South Korea, emphasized that "the UN CCW is the most appropriate forum for discussions on LAWS."[12] Thirty-nine countries from Latin America and Europe, including Austria, Chile, France, Germany, Italy, Pakistan, Spain, Uruguay, and Brazil, also expressed their support, stating that discussions within the LAWS GGE provide a sufficient basis for the development of global LAWS governance.[13]

UN Secretary-General Guterres emphasized the need for regulations on LAWS governance within the UN CCW framework to be established by 2026, and participating states are also discussing progress toward achieving the goal of establishing a LAWS governance system based on consensus among participating states by the 7th Review Conference in 2026. Despite the shock of attacks by the United States and Israel on Iran, participating states are striving to reach an agreement and produce final results within the year, holding the first session of 2026 in Geneva from March 2-6, 2026.[14]

In the early stages of the LAWS GGE discussions, Russia was the most vocal in opposing a complete ban and regulation of LAWS. However, major powers including the United States, China, the United Kingdom, Israel, India, Australia, and Japan, particularly many technology-leading nations, also clearly expressed a preference for permitting LAWS rather than regulating them. South Korea was among these nations. Conversely, countries that were technologically lagging in the use of LAWS tended to favor a complete ban. They expressed concern about the use and proliferation of weapon systems capable of lethal force without human control. For instance, 54 African nations issued a statement in 2018 urging the establishment of a legally binding regime to prohibit the use of LAWS. Meanwhile, a number of countries, including the EU, Austria, the Netherlands, Spain, Norway, France, Germany, and Italy, pursued a two-tier approach. Their position was that LAWS must also comply with existing international law, such as international humanitarian law, and therefore, while use that violates international humanitarian law should be completely prohibited, discussions should focus on methods for strict regulation in other areas.

As the governance competition between a complete ban, complete permission, and a two-tier approach for LAWS progressed, support for the two-tier approach rapidly expanded, with an agreement in principle reached that LAWS must also adhere to existing international humanitarian law.[15] In 2019, during the adoption of 11 guiding principles on the use of LAWS by the LAWS GGE, it was clarified that international humanitarian law must apply to the use of LAWS, that human intervention and responsibility must be maintained during lethal attacks, that new technologies must comply with international legal obligations, and that actions hindering the peaceful development of technology are prohibited. Furthermore, on December 2, 2024, a resolution based on a dual approach, calling for a complete ban on the use of LAWS, was adopted by the UN General Assembly. The majority of UN member states (166 countries), including the United States, which had shown some hesitation towards the two-tier approach, along with South Korea and Japan, voted in favor.[16] International organizations such as UN Secretary-General Guterres and the ICRC also clearly supported the two-tier approach, stating that autonomous attacks without meaningful human intervention and control must be completely prohibited, and that restrictions should be placed on the targets and scope of LAWS use.

Based on this, the rolling text circulated to participating states after the second session in 2025, drafted under the name of the Chair, included the following points regarding the principle of applying international humanitarian law.[17] Firstly, the rolling text includes a provision stating that international humanitarian law fully applies to the use and development of LAWS (Article 3 in Box 2). It also clarified, based on the two-tier approach, that what is prohibited by international humanitarian law in the context of existing weapon systems and the use of force must also be prohibited for LAWS (Articles 1 and 2 in Box 3).[18] Accordingly, the rolling text includes a provision stating that "the use of LAWS that is inherently indiscriminate, causes excessive injury or unnecessary suffering, or is inherently indiscriminate in its application is prohibited under any circumstances." It also clarifies that the use of LAWS must adhere to the principles of proportionality and precaution, and that civilians or civilian property must never be targeted under any circumstances. In the first session meeting in March 2026, where feedback from participating states on this matter was gathered, the majority of participating states clearly expressed their agreement with the content of the Chair's rolling text regarding the application of international humanitarian law.[19]

Examining the flow of discussions within the LAWS GGE, it appears that significant progress has been made toward a consensus on the broad framework of LAWS governance, including the application of international humanitarian law and the two-tier approach. However, despite the principled agreement on applying international humanitarian law, contentious debates continue regarding how to apply it to regulate LAWS, specifically concerning the concrete application and compliance methods of international humanitarian law.

Many European countries, including France, Germany, Denmark, and Finland, argue that LAWS should be prohibited in areas where regulation under international humanitarian law is difficult, and advocate for clear temporal and geographical limitations on the use of LAWS.[20] The rolling text, authored by Chair Robert in den Bosch of the Netherlands, also proposes detailed conditions for the use of LAWS in compliance with international humanitarian law. It specifies that LAWS should not only be used exclusively against military objectives or that the number of LAWS should be limited, but also that key mission parameters such as the type and duration of targets and the geopolitical scope should not be autonomously altered by AI (without appropriate human judgment and control within a responsible command structure).[21] It also stipulates that the number of autonomously functioning LAWS should be limited and that they should be capable of being deactivated in a timely manner, for example, through self-destruction.[22] Furthermore, the rolling text reminds that regarding the attribution of responsibility for LAWS, responsibility for acts of warfare under international humanitarian law can be imposed on states, parties to armed conflict, and individuals, but not on weapon systems,[23] which clarifies that responsibility for the consequences of LAWS cannot be attributed to AI.

Conversely, the United States, South Korea, Australia, Canada, and others, while agreeing in principle to the application of international humanitarian law to the use of LAWS, expressed concerns that the international community must find a balance between technological innovation and military necessity, and that excessive regulation could lead to undue restrictions on LAWS-based defense systems.[24] Additionally, while they generally agree to prohibit the use of LAWS targeting civilians in accordance with international humanitarian law, they expressed concern about the condition of prohibiting the use of LAWS targeting civilians "under any circumstances," proposing adherence to Article 51 of Additional Protocol I to the Geneva Conventions, which states that civilians shall enjoy general protection from the dangers arising from military operations. They also argued for the deletion of the provision regarding the limitation of the quantity of LAWS used, emphasizing that such limitations are not in compliance with international humanitarian law. China also appears to be trying to prevent excessive regulation of AI LAWS, stating that overly detailed regulations would weaken the possibility of reaching consensus on LAWS governance.[25]

The divergence of opinions among major powers regarding the scope and method of applying international humanitarian law leads to a gap among these nations concerning the necessity of concluding a new treaty on LAWS, one of the core objectives of the LAWS GGE. Many European countries, as well as African and Global South nations, advocate for the conclusion of a new treaty on LAWS or a convention prohibiting LAWS, as this would allow for the legalization of a significant level of regulation on the use and development of LAWS. In contrast, the United States, China, Russia, Israel, South Korea, and Japan are opposed to concluding a separate international treaty on LAWS. Notably, the United States maintains that existing international humanitarian law is sufficient for regulating LAWS governance. This aligns with the U.S. Department of Defense's emphasis on "lawful use" rather than "just use" in the military application of AI. If the very existence of 'law' as a basis for determining illegality is absent, the statement that AI's military use will be restricted only when it violates the law, not norms, implies a desire to minimize constraints and limitations on the military use of AI, including AI LAWS.

Consequently, the current competition in global LAWS governance exhibits less of a US-China rivalry and more of a clash of norms between technology-leading nations and less developed countries. Amidst the belief and sense of crisis that AI LAWS will provide a decisive advantage in the future military security order,[26] technology-leading nations are reluctant to accept even the minimal constraints of applying international humanitarian law, while many developing and less developed countries strongly desire the conclusion of a separate LAWS treaty or prohibition convention to mitigate the LAWS gap (similar to how the EU focused on normative power after the Cold War). Furthermore, the gap in positions regarding the conclusion of a new treaty on LAWS is manifesting as a competition of norms surrounding detailed aspects such as the definition of LAWS and human intervention and control.

III. Key Issues in the LAWS GGE

In the first session meeting of 2026, held in March 2026, participating states maintained their agreement on the principle of applying international humanitarian law and decided to continue efforts toward the goal of concluding a new legally binding treaty on LAWS.[27] However, significant gaps still exist on key issues such as the definition of LAWS and the necessity and degree of control through human judgment. This paper will focus on the issues of defining LAWS and human judgment and control to analyze the dynamics of governance competition between technology-leading nations like the United States and China, and the EU and Global South countries.[28]

1) Definition of LAWS

One of the key issues in the LAWS GGE is the definition of LAWS, which serves as the starting point for discussions on LAWS governance. This is because the definition of LAWS remains unclear, despite the fact that the manner and scope of applying general principles governed by international humanitarian law can vary significantly depending on what is considered LAWS. At the first session meeting of 2026, two particular issues became points of contention regarding the definition and core characteristics of LAWS: one is whether the function of identification should be included in the core characteristics of LAWS, and the other is how to define 'lethality'.

First, regarding whether the definition of LAWS should include the function of target identification, this issue is closely linked to the core elements of international humanitarian law, namely the protection of civilians and the distinction between combatants and civilians (or military objectives and civilian objects) for that purpose. It is also a practical issue, given that automatic target identification functions using AI are already being employed on actual battlefields, such as Israel's Lavender system.

A number of countries argued that identification should be excluded from the definition of LAWS, particularly when defining the characteristics of autonomy. Their position was that if the autonomous process or characteristics of LAWS are described as "identify-select-engage," then identification is a prerequisite for selection and an inherent function within the selection process, making separate explicit mention unnecessary.[29] Other countries argued for the exclusion of identification for entirely different reasons. They pointed out that including identification in the definition of LAWS could have the counterproductive effect of overly restricting the scope of LAWS governance and could create potential loopholes.[30] For example, there is a risk of arguing that a system is not LAWS if the identification process, or part thereof, is performed by a human.

Conversely, EU countries such as Germany and France, along with South Korea and Japan, presented opposing views, arguing that identification and selection are distinct functions and stages, and maintained the position that identification should be included in the definition of LAWS.[31] This is because including identification in the process by which LAWS attack targets is clearer and enhances consistency with international humanitarian law, such as the principle of distinguishing between combatants and non-combatants.[32] Furthermore, they expressed concern that if the function of identification is not included in the definition of LAWS, the distinction from other weapon systems could be weakened.

Amidst the differing opinions on the inclusion of identification in the definition of LAWS, the rolling text drafted in May 2025, reflecting the outcomes of the first session meeting in March 2025, proposed defining LAWS as "an integrated combination of weapon systems and technological components that can select and engage targets without user intervention."[33] This definition omitted the 'identification' function, which had been included in previous definitions of LAWS. The Chair assessed this change as the most significant development regarding the definition of LAWS from the first session meeting of 2025.[34] It was also emphasized that the exclusion of identification was an effort to minimize potential loopholes in the LAWS treaty while avoiding an overly restrictive approach.[35]

However, the rolling text dated December 18, circulated after the second session meeting in 2025, re-included identification in the definition of LAWS.[36] Additionally, as a measure to address concerns that the inclusion of identification might create potential loopholes, a sub-clause was added stating that "cases where the target parameters are set by a human" are also included as LAWS.[37]Although China and others expressed opposition to these annexes, a majority of countries, including the United States, agreed on the necessity of adding annexes. Discussions and revisions of these annexes continued during the first session in 2026, but an agreement appears to have been reached on a direction that includes identification under conditional terms.[38]

In contrast, the issue of including 'lethality' remains a core point of contention among major countries. This is because it determines whether the regulations and governance discussed in the LAWS GGE should target only autonomous weapon systems (AWS) that possess or cause lethal capabilities, or all forms of autonomous weapon systems. If the former, systems like missile defense systems, close-in weapon systems, and active protection systems could be excluded from regulation, making the issue even more sensitive. The question of what degree of killing or destruction constitutes 'lethal' is also a point of significant disagreement.

EU countries such as Austria and Switzerland have clearly stated their opposition to including the adjective 'lethal' in the definition of LAWS. They argue that since the concept of 'lethality' does not exist in existing international humanitarian law, it should not be imposed as a condition in LAWS governance, which is based on the application of international humanitarian law.[39] Furthermore, they point out that lethality is an effect of weaponry and a result of engagement, not a characteristic of the weapon system itself. They also expressed concern that including lethality could lead to the emergence of LAWS that fall outside the scope of governance.[40]

Conversely, the United States, China, Russia, and South Korea argue that 'lethality' should be included. Russia emphasized that lethality has been a core characteristic since the inception of the UN CCW GGE on LAWS in 2016, clarifying the subject of discussion and governance, while China stressed that the lethality requirement should be explicitly stated in the rolling text. The United States also emphasized that the mandate given to the LAWS GGE pertains to lethal autonomous weapon systems. There is also a conflict between the preferences of the US, China, Russia, and others, and the preferences of the EU regarding the definition of lethality. First, the Chair's rolling text, circulated in May 2025, defined lethality as 'the capacity to cause death to a person'.[41] While there may be cases of injury or destruction of objects that do not result in death, the explanation suggests focusing on the capacity to cause death rather than merely the outcome. This definition of lethality was assessed as a constructive effort to bridge the gap between those advocating for the prohibition and regulation of all autonomous weapon systems and those arguing for the regulation of autonomous weapon systems only in cases directly linked to killing.[42]

However, the United States argues that the term should be changed from 'lethality' to 'with lethal force.' For example, the US emphasized that 'autonomy' should be understood based on function or mission, regardless of whether specific technologies or capabilities enabling autonomous functions are included, and that autonomy is ultimately a capability of the weapon system, thus differing from indiscriminate weapons that are prohibited by existing international humanitarian law.[43] In other words, their position is to focus on the outcome of the act rather than the properties or capabilities of the weapon system. South Korea and India have also expressed agreement with the US's opinion.

Reflecting these opinions, the most recently circulated rolling text of December 2025 included the term 'lethal' in the definition of LAWS under discussion, while also adding a new annex (Article 1, Paragraph B): 'LAWS shall not be excluded from the category of LAWS even if they can be used in ways that do not result in the taking of human life, such as the destruction or damage of objects or causing injury to persons.' This is an attempt to reflect the preferences of countries opposing the inclusion of lethality, such as the EU. Indeed, countries like Switzerland, Brazil, and Ireland expressed the view that the newly added annex is a realistic compromise.[44] The United States proposed that the annex is unnecessary.

2) Human Judgment and Control

The necessity and scope of meaningful human control and judgment have also been subjects of intense debate within the LAWS GGE.[45] Many countries, including Brazil and Mexico, emphasize the importance of human control and argue that it should be included as a separate, distinct provision to ensure compliance with international humanitarian law and clarity in accountability, thereby making it a core requirement for LAWS governance treaties. Germany, Austria, and others argued that for international humanitarian law to function effectively in the age of AI, human judgment must be involved in the use of force. The EU also emphasized that human judgment and control should apply throughout the entire life cycle of LAWS, including development, deployment, and use. Furthermore, they criticized that the discussions in the GGE on LAWS have resulted in excessively low-level compromise texts to accommodate the diverse opinions of various countries, and argued that the rolling text, which serves as the starting point for discussions, should not be further weakened.[46] International organizations such as the UN and the International Committee of the Red Cross (ICRC) also stressed that human control should be explicitly stated as a core function of LAWS, arguing that in addition to temporal and spatial limitations (such as specifying clear periods of use and geographical areas), there must be the ability for humans to interrupt or deactivate LAWS in emergencies.

Conversely, the United States, China, and Japan clearly stated their opposition to explicitly including human control as a separate provision, arguing that it is merely a means to ensure compliance with international humanitarian law. The US is negative about explicitly including conditions for human control in a way that could create legal obligations. They also suggested that human judgment cannot always be assumed to be superior to autonomous judgment by AI, and that autonomous judgment by AI might be more effective in fulfilling the objectives of international humanitarian law, such as protecting civilians.[47]

In the current session, the issue of human intervention and control over LAWS is being debated regarding the inclusion and wording of 'context-appropriate human judgment and control.' Proponents of explicit inclusion emphasize that the phrase 'contextually appropriate human judgment and control' is necessary for LAWS to adhere to core principles of international humanitarian law, such as the principles of distinction, proportionality, and precaution in attack. They explain that the phrase 'contextually appropriate human judgment and control' can serve as a mechanism to bridge the gap between the positions of major countries regarding the degree of human control over LAWS.[48] For example, European countries such as Bulgaria, Denmark, France, and Germany emphasized that there is a growing consensus on the importance of the 'human element' in LAWS discussions, and argued that expressions previously used in LAWS discussions, such as 'sufficient,' 'adequate,' and 'meaningful,' can be encompassed by the phrase 'context-appropriate human judgment and control.'[49]

Furthermore, they explain that 'context-appropriate' refers to the level and type of human intervention and control, considering factors such as the type of weapon, planned mission, and operational environment and conditions. Human judgment, in this context, refers to the exercise of judgment by humans when initiating hostilities in accordance with international humanitarian law, including the principles of distinction, proportionality, and precaution in attack, during the development and use of LAWS.[50]

In response to these arguments, the United States emphasized that 'context-appropriate human judgment and control' cannot necessarily be considered meaningful human control, and that the term 'appropriate' is an ambiguous standard that cannot determine compliance with international humanitarian law. They also clarified their opposition to interpreting the prohibition of LAWS without context-appropriate human judgment and control as a ban on all autonomous weapon systems that operate autonomously after deployment without human intervention, such as fire-and-forget systems.[51] Furthermore, the United States suggested that the term 'operator' is more appropriate than 'human user.'[52] They explained that this clarifies the nature of direct human control over weapon systems and is a military and legal term. Additionally, the US and others argued that the term 'human' should not be used, whether in 'human operator' or 'human user.' They explained that international humanitarian law also uses the term 'user' without including the term 'human,' as it is understood that the user is naturally human. They also expressed concern that using 'human user' could imply the existence of non-human users.[53] Japan expressed concern that context-appropriate human judgment and control should be a means to use LAWS in accordance with international humanitarian law, rather than a condition in itself.[54] South Korea presented similar opinions to the US and Japan, expressing a preference for the previous wording, 'human judgment and control are necessary to comply with international humanitarian law.'[55] Furthermore, China also expressed concern about the need to clarify objective criteria for what constitutes 'appropriate.'[56]

Amidst these concerns, regarding the criteria for 'context-appropriate,' some countries argue that legal considerations, as well as ethical considerations, should be explicitly included in the chain of command for AI LAWS. For example, Brazil, Peru, and Cuba maintain that ethical considerations must be included as they align with the guiding principles and the Martens Clause of the 2023 GGE report. In response, Japan and Israel argued for the deletion of this phrase, pointing out that there is no universally agreed-upon ethical standard for 'ethical considerations' among states.[57]

IV. Prospects and Implications of LAWS Governance

LAWS governance is a critical and urgent matter, as it is one of the key variables in the development of the future world order and establishes the framework for global governance concerning the military use and development of rapidly advancing AI. To this end, the current LAWS GGE is accelerating its discussions to finalize the outcome by 2026.[58] However, despite meaningful progress, including principled agreement on the development and use of LAWS in compliance with international humanitarian law and finding common ground on the definition of LAWS, it appears unlikely that the LAWS GGE will achieve its intended final outcome. In particular, the possibility of concluding a legally binding instrument, such as a treaty or a prohibition convention as an annexed protocol to the UN CCW, which many countries prefer, is not high.

The first reason is that the UN CCW requires the consensus of all States Parties. Currently, the gap between the positions and preferences of major countries on key issues of the LAWS GGE, such as the inclusion of lethality in the definition of LAWS, the method of human control, and the prohibition of LAWS development, remains significant. However, the opportunities for coordination are limited to the second session in 2026, scheduled for the first week of September, and a few informal consultations to be held under the Chair's leadership before then. Therefore, it is expected to be difficult to reach an agreed text that all States Parties can accept.

The second reason is that countries leading in AI technology, such as the United States and China, do not desire a LAWS treaty. In particular, the US's negative perception and preference regarding global LAWS governance are intertwined with its self-destructive approach towards international organizations, international law, and universal human rights norms, which includes criticism and rejection of the rules-based liberal international order it once led, weakening institutional support, reducing financial contributions, and losing normative leadership.[59] This also contributes to growing concerns about the depalawization of international politics and the crisis of norms, where the influence of law and principles, including international law, is weakening in international relations.[60] Nevertheless, even if the possibility of reaching a final agreement within the LAWS GGE is low, understanding and analyzing the discussion process of the LAWS GGE is necessary for two reasons. The first is that global governance must always be understood with consideration for change. In other words, one must examine the dynamics of which actors advocate for which values and norms, how they publicize them, and through relationships with which actors they attempt to have their preferred norms accepted and institutionalized in the international community. The second reason is that discussions within the LAWS GGE may continue outside the framework of the UN CCW. The Convention on the Prohibition of Anti-Personnel Mines was concluded through the Ottawa Process, initiated by major countries and non-governmental actors, including Canada, who were disappointed with the slow progress of discussions within the UN CCW, and was later adopted as the Second Protocol to the CCW. Similarly, the Convention on Cluster Munitions was concluded after major countries and non-governmental actors, including Norway and New Zealand, met in Oslo to adopt a declaration and initiate discussions outside the CCW framework, as discussions within the UN CCW showed little progress. It is highly possible that LAWS governance will proceed in a similar manner. In particular, considering that both the Ottawa Convention and the Oslo Convention were concluded relatively quickly, within about a year, after discussions that had been progressing slowly within the CCW moved outside of it, preparations for discussions outside the LAWS GGE framework within this year are also necessary.

Therefore, what should be South Korea's approach and strategy for the LAWS governance discussions that may emerge outside the CCW framework? First, it must align with South Korea's preferences and strategies for AI governance. Second, realistically, South Korea needs to participate in the AI ecosystem led by the United States. Furthermore, given the reality of a divided nation and an increasingly pressured security environment with escalating arms races, South Korea also has a significant need for the use of military AI, including LAWS. At the same time, the transition in the global security order and normative system is becoming more pronounced due to the potential for prolonged changes in the policies of allies and friendly nations towards the US. Solidarity and cooperation with like-minded countries are also crucial for overcoming the crisis of this transitional period, thus the strategy of middle power normative diplomacy should not be abandoned. South Korea, while not a signatory to the Ottawa Convention due to the specific circumstances of the Korean Peninsula, expressed its principled support for the prohibition of anti-personnel mines as the norm was being established through the Ottawa Process. Therefore, South Korea must actively participate in the LAWS governance discussions, both within and outside the LAWS GGE.[61] That is, we must examine the dynamics of which actors advocated for certain values and norms, how they publicized them, and through relationships with which actors they attempted to have their preferred norms accepted and institutionalized in the international community. The second reason is that discussions within the LAWS GGE may continue outside the framework of the UN CCW. The Ottawa Process, which led to the Anti-Personnel Mine Ban Convention, was concluded when major countries and non-governmental actors, disappointed by the sluggish progress of discussions within the UN CCW, gathered in Ottawa to begin discussions separately from the CCW, and it was subsequently incorporated as Protocol II of the UN CCW. Similarly, the Convention on Cluster Munitions was concluded after major countries and non-governmental actors, including Norway and New Zealand, gathered in Oslo to adopt a declaration and begin discussions outside the CCW framework due to a lack of progress in the UN CCW. It is highly probable that LAWS governance may proceed in a similar manner. In particular, considering that both the Ottawa Convention and the Oslo Convention were concluded within a relatively short period of about one year after discussions, which had been proceeding at a somewhat slow pace within the CCW, began outside the CCW, it is necessary to prepare for the possibility that discussions may proceed outside the LAWS GGE framework within this year.

What, then, should be South Korea's approach and strategy for the LAWS governance discussions that may emerge outside the CCW framework? First, it must align with South Korea's preferences and strategies regarding AI governance. Furthermore, realistically, South Korea needs to participate in the AI ecosystem led by the United States. Additionally, consideration must be given to the reality of the Korean Peninsula being a divided nation and the increasing pressure of the arms race in the security environment, which necessitates the use of military AI, including LAWS, for South Korea. Concurrently, we must consider the growing transformative changes in the global security order and normative frameworks, such as the prolonged possibility of policy changes by the United States towards its allies and friendly nations. In overcoming the crisis of this transitional period, solidarity and cooperation with various like-minded countries are also crucial, thus the strategy of middle power normative diplomacy should not be abandoned. South Korea, in the past, did not become a signatory to the Anti-Personnel Mine Ban Convention, considering the specific circumstances of the Korean Peninsula, even as the norm against anti-personnel mines was being legislated through the Ottawa Process, but it did express its principled agreement with the prohibition of anti-personnel mines. Therefore, South Korea must actively participate in the LAWS governance discussions taking place within and outside the LAWS GGE.■

[1]Seol, Inhyo. 2026. "Analysis of the Development Trends of AI Military Innovation: Two Perspectives on the Speed of Innovation and Cases of the US and China." EAI Working Paper, January 27.

[2]Ramkumar, Amrith, Keach Hagey, and Vera Bergengruen. 2026. "Pentagon Used Anthropic’s Claude in Maduro Venezuela Raid."Wall Street Journal, February 15. https://www.wsj.com/politics/national-security/pentagon-used-anthropics-claude-in-maduro-venezuela-raid-583aff17.(Accessed: April 5, 2026).

[3]Amaral, Nilza. 2026. "The Iran War Highlights the Creeping Use of AI in Warfare."Chatham House. March 27. https://www.chathamhouse.org/2026/03/iran-war-highlights-creeping-use-ai-warfare.(Accessed: April 5, 2026); Brown, Michael. 2026. "The First AI War: How the Iran Conflict is Reshaping Warfare."Forbes. March 30. https://www.forbes.com/sites/mikebrown/2026/03/30/the-first-ai-war-how-the-iran-conflict-is-reshaping-warfare/ .(Accessed: April 5, 2026); Hammad, Noor. 2026. "The Proliferation of AI-enabled Military Technology in the Middle East."IISS Charting Middle East. April 2. https://www.iiss.org/online-analysis/charting-middle-east/2026/04/the-proliferation-of-ai-enabled-military-technology-in-the-middle-east/.(Accessed: April 5, 2026).

[4]Frenkel, Sheera and Julian E. Barnes. 2026. "Defense Dept. and Anthropic Square Off in Dispute Over AI Safety."New York Times, February 18. https://www.nytimes.com/2026/02/18/technology/defense-department-anthropic-ai-safety.html.(Accessed: April 5, 2026).

[5]Ramkumar, Amrith. 2026. "Trump Administration Shuns Anthropic, Embraces OpenAI in Clash Over Guardrails."Wall Street Journal, February 27. https://www.wsj.com/tech/ai/trump-will-end-government-use-of-anthropics-ai-models-ff3550d9.(Accessed: April 5, 2026).

[6]Metz, Cade, Julian E. Barnes, and Sheera Frenkel. 2026. "Pentagon Officially Notifies Anthropic it is a 'Supply Chain Risk'."New York Times, March 5. https://www.nytimes.com/2026/03/05/technology/anthropic-supply-chain-risk-defense-department.html .(Accessed: April 5, 2026).

[7]Dorsey, Jessica, Elke Schwarz, Ingvild Bode, Zena Assaad, and Neil Renic. 2026. "The Pentagon/Anthropic Clash over Military AI Guardrails."Opinio Juris. February 26. https://opiniojuris.org/2026/02/26/the-pentagon-anthropic-clash-over-military-ai-guardrails/.(Accessed: April 5, 2026).

[8]Dorsey, Jessica, Elke Schwarz, Ingvild Bode, Zena Assaad, and Neil Renic. 2026. "The Pentagon/Anthropic Clash over Military AI Guardrails."Opinio Juris. February 26. https://opiniojuris.org/2026/02/26/the-pentagon-anthropic-clash-over-military-ai-guardrails/.(Accessed: April 5, 2026)

[9]As AI is now a core element in the discussion of LAWS governance, the outcomes of global LAWS governance also provide an opportunity to examine the landscape of global AI governance.

[10]Moon, Yong-il. 2024. "Convergence or Isomorphism?: A Comparative Analysis of Data Privacy Legislation in the EU, the US, and China." International Area Studies Review 28 (2).

[11]The UN Office of Disarmament Affairs addresses new challenges including AI in science and military domains, armed uncrewed systems, ICT security, space, missiles, and LAWS along with civil society and arms control. See the UN Office of Disarmament Affairs website. https://disarmament.unoda.org/en. (Last accessed: April 5, 2026).

[12]UN CCW/GGE.1/2025/WP.7 (Joint submission by the United States, et al.).

[13]UN CCW/GGE.1/2025/WP.8 (Joint statement by 39 countries).

[14]See the Chair's opening remarks during the afternoon session on March 7, 2026. Videos of the meetings held during the first session are also available on the website. https://meetings.unoda.org/ccw-/convention-on-certain-conventional-weapons-group-of-governmental-experts-on-lethal-autonomous-weapons-systems-2026.(Last accessed: April 5, 2026)

[15]Perrin, Benjamin. 2025. "Lethal Autonomous Weapons Systems & International Law: Growing Momentum Towards a New International Treaty."American Society of International Law Insights 29 (1).

[16]Only three countries—Russia, Belarus, and North Korea—clearly expressed their opposition to the resolution. Meanwhile, China was one of the 15 countries that abstained from the vote on the resolution. This indicates a negative stance towards discussions on regulating LAWS. The countries that abstained from voting on the resolution, in addition to China, were Estonia, Fiji, India, Iran, Israel, Latvia, Lithuania, Nicaragua, Poland, Romania, Saudi Arabia, Syria, Turkey, and Ukraine (UN General Assembly Resolution A/RES/79/62. December 2, 2024).

[17]UN CCW/GGE.1/2025 (Revised rolling text as of May 12, 2025).

[18]UN CCW/GGE.1/2025.

[19]The main points of the first session of the UN CCW GGE on LAWS, held in early March 2026, can be found in the Chair's summary released on April 1st {UN CCW GGE.1/2026.WP.2 https://docs-library.unoda.org/Convention_on_Certain_Conventional_Weapons_-Group_of_Governmental_Experts_on_Lethal_Autonomous_Weapons_Systems_(2026)/CCW-GGE.1-2026-WP.2.pdf .(Last accessed: April 5, 2026).

[20]UN CCW/GGE.1/2025/WP.4.

[21]UN CCW/GGE.1/2025/WP.9, Box 3.

[22]UN CCW/GGE.1/2025/WP.9, Box 3.

[23]UN CCW/GGE.1/2025/WP.9, Box 3.

[24]UN CCW/GGE.1/2025/WP.6.

[25]Afternoon session on March 3, 2026. https://webtv.un.org/en/asset/k1y/k1ywe5vjyq.(Last accessed: April 5, 2026)

[26]Garcia, Eugenio. 2024. "Conclusions: Charting the Challenge of AI IR."International Studies Review 26 (2); Cha, Tae-seo. 2026. “The AI Revolution and Republican Security Theory: The Resurgence of the Dual Dilemma of Anarchy and Hierarchy,” 『EAI Working Paper』, February 3.

[27]Afternoon session on March 7, 2026. https://webtv.un.org/en/asset/k1m/k1m35gbktf.(Last accessed: April 5, 2026)

[28]Another contentious issue during the first session of 2026 was the prohibition of the development of LAWS. The current rolling text addresses the prohibition of the employment of LAWS, and the core issue became whether to replace this with a prohibition on the development of LAWS. Countries advocating for a ban at the development stage, such as Brazil, Mexico, Norway, Pakistan, and Cuba, argue that they must follow the precedent of existing treaties that prohibit not only the use but also the development of inhumane weapons, such as the Biological Weapons Convention (BWC), which entered into force in 1975, and the Chemical Weapons Convention (CWC), which entered into force in 1997. Conversely, major countries like the United States, South Korea, Japan, Israel, the United Kingdom, and Australia argue that international humanitarian law inherently regulates 'use,' and therefore, it should be followed, pointing to issues of ambiguity in the meaning and scope of development. China has taken the position that it should be left to national discretion, and the EU has not shown a consistent stance, with France supporting the prohibition of development while countries like Germany and Sweden express conditional opposition.

[29]UN CCW/GGE.1/2025/WP.9. (Rolling text as of May 12, 2025).

[30]UN CCW GGE on LAWS Rolling text-December 18, 2025 version. p.3.

[31]UN CCW GGE on LAWS Rolling text-December 18, 2025 version. p.3. The positions of each country can be confirmed in the video of the first session meeting.

[32]UN CCW/GGE.1/2026/WP.2.

[33]"an integrated combination of one or more weapons and technological components, that can select and engage a target, without intervention by a human operator in the execution of these tasks." {UN CCW/GGE.1/2025/WP.9 (Rolling text as of May 12, 2025)}.

[34]UN CCW/GGE.1/2025/WP.9. (Rolling text as of May 12, 2025).

[35]UN CCW/GGE.1/2025/WP.9. (Rolling text as of 12 May 2025).

[36]UN CCW GGE on LAWS Rolling text-18 December 2025 version

[37]Article 1(A) states: “The fact that a human operator sets the parameters of the target(s) does not exclude the system from the characterization.”

[38]UN CCW/GGE.1/2026/WP.2.

[39]See statement by Austria during the afternoon meeting on March 2, 2026.

[40]See statement by Ireland during the morning meeting on March 2, 2026.

[41]UN CCW/GGE.1/2025/WP.9. (Rolling text as of 12 May 2025).

[42]UN CCW/GGE.1/2025/WP.9. (Rolling text as of 12 May 2025).

[43]UN CCW/GGE.1/2025/wp.6.

[44]See morning meetings on March 2, 2026 (Ireland and Switzerland) and the morning meeting on March 5, 2026 (Brazil).

[45]In the early discussions on governance of autonomous weapon systems, there were strong calls to ban LAWS that target humans. The argument was that the determination of human life based on mechanical calculations and judgments, excluding moral and ethical decision-making, fundamentally infringes upon human dignity. However, the prohibition of anti-personnel systems, such as so-called 'killer robots,' is currently absent from the main agenda items of the LAWS GGE. This is a point of strong criticism from civil society and other groups.

[46]See opinions of the EU, Austria, and Germany during the morning meetings on March 2 and March 3, 2026.

[47]See opinions of the United States during the morning meetings on March 3 and March 4, 2026.

[48]UN CCW/GGE.1/2025/WP.9.

[49]UN CCW/GGE.1/2025/WP.4.

[50]UN CCW/GGE.1/2025/WP.4.

[51]See intervention by the United States during the afternoon meeting on March 4, 2026.

[52]For example, the United States proposed that an operator refers to a human who can activate, control, or monitor LAWS, and that a commander who merely issues orders should not be considered an operator (UN CCW/GGE.1/2025/WP.6; see statement by the United States during the morning meeting on March 3, 2026).

[53]See statement by the United States during the morning meeting on March 3, 2026.

[54]See statement by Japan during the morning meeting on March 3, 2026.

[55]See statement by the Republic of Korea during the morning meeting on March 3, 2026.

[56]See statement by China during the morning meeting on March 2, 2026.

[57]See morning and afternoon meetings on March 3, 2026.

[58]The Chair indicated that a revised rolling text, focusing on the outcomes of this first session, would be circulated to all states in April, and discussed during the second session scheduled from August 31 to September 4, 2026. The Chair also clarified that informal consultations would be held between sessions to facilitate agreement. See the Chair's statement during the afternoon meeting on March 7, 2026.

[59]Cordell, Rebecca and Alex Dukalskis. 2025. "Authoritarianism, Global Politics, and the Future of Human Rights."International Organization 79; Reeder, Bryce. 2025. "The Future of United Nations Peacekeeping in a Fragmenting World."International Organization 79.

[60]For instance, the United States declared that it would not undergo the UN Universal Periodic Review, a groundbreaking development and outcome of the international human rights regime, for the first time in the world in 2025. Consequently, the UPR for the United States was exceptionally postponed until the latter half of 2026 {Human Rights Watch. 2025. “US Skips UN Periodic Rights Review.” November 7. https://www.hrw.org/news/2025/11/07/us-skips-un-periodic-rights-review.(Last accessed: April 5, 2026).

[61]Avant, Deborah, Martha Finnemore, and Susan Sell, eds. 2010. Who Governs the Globe? Cambridge Studies in International Relations. Cambridge: Cambridge University Press.

■Author: Moon Yong-il_Professor, Department of International Relations, University of Seoul.

■Editor: Lim Jae-hyun_Research Fellow, EAI

    Inquiries: 02 2277 1683 (ext. 209) | jhim@eai.or.kr