Symposium on Cyber, International Law and Artificial Intelligence in Armed Conflict: Piercing the Veil – Why Siloed AI Governance Is Untenable

[Jimena Sofía Viveros Álvarez is a senior international lawyer and peace and security expert, CEO of IQuilibriumAI, OECD and UNESCO AI Expert, and former Member of the UNSG HLAB and GC-REAIM Commissioner]

Contemporary armed conflicts defy the boundaries that construe conservative legal frameworks, and as the threshold has de facto been lowered, distinguishing between war and peace has become increasingly challenging. In light of this, a basic premise is that the whole international corpus juris applies to artificial intelligence (AI) as it is “technology neutral”; thus, International Humanitarian Law (IHL), jus ad bellum, International Human Rights Law (IHRL), International Criminal Law (ICL), and State responsibility are still apposite, notwithstanding that some gaps persist for an AI governance lex specialis. Indeed, United Nations General Assembly (UNGA) Resolution 79/239 affirmed that international law, including the UN Charter, IHL, and IHRL, applies to AI throughout its lifecycle. As UN Secretary-General (UNSG) António Guterres has manifested, the integration of civilian AI applications into the military, and AI’s inherent repurposable nature, are blurring the lines between the traditional “military” and “civilian” domains, both operationally and legally. While interpreting existing norms to girdle the emergence and ubiquity of these technologies is correct and necessary, the author holds that it is also true that not all uses, applications, and emerging technological capabilities are encompassed by existing provisions, however much we attempt to stretch them out for this purpose. The author argues that AI’s inherent dual-use, foundational, general-purpose, and ubiquitous nature defies the status quo of orthodox governance structures, which artificially dissociate military and civilian domains as if they were mutually incompatible, while in fact no clear-cut dissection can be made between ambits, rendering this segregation untenable.

Dual-Use and Paradigm Shifting

The Human Rights Council Advisory Committee defined “dual-use technologies” as:

“innovations with both civilian and military applications, with potential uses in the commercial, public and military domains.”

AI’s intrinsic dual-use nature embodies these characteristics to an unprecedented degree, fading boundaries between civilian and military domains and applications. While AI development is predominantly commercial or “civilian,” its myriad applications pervade military and security spheres throughout its lifecycle, i.e., from its design, development, use, and decommissioning, regardless of whether AI systems were initially conceived for civilian or military purposes, creating regulatory gaps and enabling “forum shopping” to evade oversight. Simultaneously, commercial technologies are increasingly being used for military purposes; for instance, in Russia’s aggression war, the SpaceX Starlink satellite constellation, typically providing internet broadband to civilians, was deployed to support Ukrainian military communications, and generative AI (GenAI) is used to assist decision making, cybersecurity and intelligence operations. Anthropic’s AI model, Claude, was reportedly used by the United States (US) during Maduro’s capture operation in Venezuela. Unlike previous disruptive technologies that remained within government’s control, AI’s development is driven predominantly by industry. Moreover, AI’s open-source characteristics enable instantaneous cross-border transfers, ease proliferation to non-State actors, and make it inherently repurposable, which exacerbates risks, especially of misuse.

Autonomy, in particular, cannot be regarded as simply another technological advancement, let alone as an “evolved” means and methods of warfare, but rather as a true paradigm shift drastically transfiguring the nature of armed conflict by the unparalleled erosion of human control over critical functions in the targeting and engagement cycles. This compendium of permutations demands corresponding transformative governance approaches, because conservative frameworks fail to grasp AI’s unique dual-use characteristics across the lifecycle of these technologies, affecting civilians and militaries alike.

Governments and Industry Have Pierced the Veil

It is the author’s view that this separation no longer exists in practice as these lines are already extremely blurred — not by governance proposals, but by the technology itself, modern-day conflicts, and State and non-State actors’ conduct, including the private sector. As noted, governments and industry have vehemently pierced the veil between civilian and military uses. Countries like China have embraced this “military-civil fusion” at least since 2014; in 2024, Germany’s Federal Ministry of Education and Research proposed revising the constitutional policy barring universities from conducting military-related research; and the European Union has “encouraged the incorporation of civilian technologies into military systems.” Alarmingly, the US has bolstered its civil-military integration, launching the Genesis Mission in November 2025; in December 2025, an Executive Order establishing a national framework for AI calls for “[US] AI companies [to] be free to innovate without cumbersome regulation,” aligned with the US AI Action Plan from July 2025. The Genesis Mission has been explicitly framed as urgent as the Manhattan Project, with the “dominance” magniloquence permeating US official documents mandating accelerated development explicitly at the expense of safety, legal, and ethical restraint.

The exponential militarization of technology is undeniable. Leading AI companies — including Google, Meta, and OpenAI — lifted restrictions on military use of their systems, enabling direct collaboration with defense agencies; and Anthropic changed its pledge to pause the scaling or delay the deployment of new AI models. In June 2025, the US military appointed executives from Palantir, Meta, OpenAI, and Thinking Machines Lab as Lieutenant Colonels in its “Executive Innovation Corps,” while the Pentagon labeled Anthropic a “supply chain risk” for not caving into threats to drop its safety measures. Technological giants have thereby gained de facto quasi-autonomous standing themselves:

“due to their control over critical digital infrastructures, data flows, compute power, human talent, and algorithmic governance.”

Framework Silos

There are two main approaches to AI governance. The first is conventional siloed governance, maintaining “a strict division between military and civil uses of technologies.” The second recognizes AI’s dual-use and transboundary nature and seeks to integrate responses across the broader peace and security domains. Most current efforts remain siloed. Following the recommendations of the UNSG High-level Advisory Body on AI, the General Assembly established the Independent International Scientific Panel on AI and the Global Dialogue on AI Governance in August 2025, yet their mandates are explicitly “limited to the non-military domain and do not refer to artificial intelligence for military purposes.” The EU AI Act explicitly states that AI systems used exclusively for military, defense or national security purposes are out of its scope while also recognizing dual-use AI systems; this carve-out has been controversial as it avails inconsistent oversight for security or surveillance AI uses. The ASEAN Guide on AI Governance and Ethics applies to organizations developing AI “in commercial and non-military or dual-use applications,” yet “dual-use applications” are not defined or provisions stipulated for them, opening a lacuna. On the military track, since 2016 the Group of Governmental Experts (GGE) on lethal autonomous weapons systems (LAWS) within the Convention on Certain Conventional Weapons (CCW) has accrued legitimacy and significant specialization, but its lack of progress has been criticized as a “road to nowhere,” and the CCW’s consensus rule has been used as a “veto power.”

Excluding dual-use considerations creates a regulatory landscape deliberately tone deaf to contextual and technological realities, rolling out dangerous blind spots precisely where autonomous systems would most likely be deployed — the gray zones between war and peace. Maintaining the segregation of isolated legal regimes is therefore implausible, as they are intrinsically convergent within the broader peace and security realm, requiring adaptive and reality-responsive global governance. The core protections afforded by IHL and IHRL are not mutually exclusive but rather cumulative, especially in murky operational contexts, abiding by the overarching foundational tenets enshrined by the Martens Clause. The proposed normative approach does not “militarize” civilian AI development or “civilianize” military operations — it simply recognizes that dual-use systems can be used for both military and non-military purposes. It also does not collapse armed conflict and peace into one undifferentiated domain; rather, it proposes to create a protective connective tissue across them, ensuring that an overarching regime can be effectively applied to technologies that operate fluidly athwart domains.

The Normative Way Forward: Red Lines

The time is ripe to move past voluntary commitments to overarching and universally binding norms enshrining core principles and red lines applicable to all AI systems across domains. This requires a two-tier approach. The first category, related to systems that pose untenable or existential risks, requires clear red-lines prohibiting fundamentally unlawful AI applications; the second seeks gradient regulatory frameworks, robust technical safeguards and ethical guardrails for systems presenting layered risks. For tier one, AI technologies posing untenable or categorical risks to foundational values should be subject to an absolute prohibition, i.e., red lines, such as: fully autonomous weapons systems (AWS) and AI decision support systems (AI-DSS) with no human oversight, unpredictable in their use of violence and thus in violation of IHL; GenAI integration into any safety-critical system; AI integrated into the research, development, authorization, and use of weapons of mass destruction; AI models with power-seeking behavior, autonomous cyberattacks, deceptive behavior, and recursive self-improvement; and AI technologies for mass surveillance.

A normative two-tier framework will require strong institutional architecture. A global AI governance body should act as a centralized, independent, and authoritative guardian institution, with the sufficient “muscle and teeth” to delineate, monitor, verify and enforce red-lines and regulations, because “verification is what turns agreements into reality.”

The imperative of red lines gained authoritative support with the UNSG’s joint call with the President of the International Committee of the Red Cross (ICRC) for the conclusion of a legally binding instrument by 2026, banning (L)AWS that “have the power and discretion to take human lives without human control,” underscoring the urgency for this red line to become enforceable. And in December 2025, UNGA Resolution 80/23 affirmed that:

“pending the total elimination of nuclear weapons, human control and oversight is maintained over command, control and communications systems of nuclear weapons, including those that integrate artificial intelligence technology.”

These developments mark welcome progress, but they risk fragmentation and lack inclusivity while the window for action is narrowing. AI threats are transboundary, concrete, immediate, and accelerating faster than governance responses.

We are not in a legal vacuum, as existing international law is technology-neutral and remains applicable to AI in armed conflict, however, what interpretation alone cannot close are the gaps and ambiguities exposed by the rapid and often unforeseeable advance of these emerging capabilities. Meeting them requires a shift from a fragmented, reactive posture to a proactive, harmonized architecture that bridges the factually artificial divide between civilian and military domains: categorical red lines for untenable risks, graduated regulation for the rest, and a specialized, universal, and centralized authority able to set norms, verify compliance, and enforce them. Such governance must be at once preventive and responsive, and inclusive of the Global South, whose exposure to these technologies’ harms is most acute. The author argues that dual-use global governance is the precondition for keeping AI’s disruption within the bounds of international peace and security.

This post forms part of the Opinio Juris symposium on International Law and Artificial Intelligence in Armed Conflict (introduced here) and draws on the author’s chapter in the forthcoming OUP volume of the same title.