The Autonomy Gap: How AI Weapons Systems Are Rewriting the Rules of Modern Conflict

At a public appearance on 28 May 2026, US Vice President JD Vance offered a candid assessment of where artificial intelligence and modern warfare intersect. "The thing I worry about most with AI is how it will change warfare," Vance said, according to a post by Disclose.tv that same day. "AI will inevitably change warfare." The remark was brief. The implications are not.

The vice president's warning landed in a media environment already saturated with drone footage from Ukraine, debate over Turkish autonomous systems in Libya and Nagorno-Karabakh, and growing tension at the United Nations over where to draw lines on machine-driven lethal force. Vance was not describing a hypothetical. He was confirming what defense planners in Washington, Beijing, and Moscow have understood for years: the integration of artificial intelligence into military systems is not a future scenario. It is the present operating environment.

The question this publication finds most consequential is not whether AI will reshape combat — the answer to that is plainly yes — but whether the institutions meant to govern armed conflict possess the will or the capacity to impose meaningful limits on systems that can select and engage targets without a human finger on the trigger.

Already in the Field

The debate over lethal autonomous weapons systems (LAWS) has long been treated as a future-shaping controversy, a matter for ethics boards and UN working groups to resolve before the technology matures. That framing is increasingly difficult to sustain. Autonomous and semi-autonomous systems are already deployed in active conflict zones, and the distinction between the two categories — systems that execute a pre-authorised targeting envelope versus systems that select targets de novo — is often less a matter of technology than of legal definition.

The Turkish Kargi loitering munition, deployed by the Azerbaijan Armed Forces in the 2020 Nagorno-Karabakh war, was described by some analysts as operating with autonomous target recognition. Israel has integrated AI-assisted targeting into its IDF operations, with the military describing these as decision-support rather than decision-making systems — a distinction critics argue is increasingly academic. The Russian military has reportedly tested the Lancet suicide drone in a manner that analysts have characterised as consistent with autonomous terminal guidance.

None of these deployments involved the science-fiction scenario of killer robots ranging freely across a battlefield. They are narrower: systems that employ machine vision and algorithmic pattern-matching to identify and strike targets within parameters defined by human operators. But the distance from that model to a fully autonomous system is algorithmic, not physical. It is a matter of threshold-setting, not fundamental redesign.

Ukraine has become the most extensively documented laboratory for AI-assisted warfare in history. The conflict has seen the deployment of thousands of commercial-grade drones repurposed for reconnaissance-strike complexes, with Ukrainian forces developing AI-assisted target identification tools to compress the sensor-to-shooter loop. Western defence suppliers have accelerated deliveries of systems described as AI-enabled. The speed of this adaptation has exceeded many early-war predictions and has provided real-world data on how human-machine teaming performs under combat stress — data that is now feeding directly into procurement decisions in NATO member states.

The structural logic driving this diffusion is not ideological. It is competitive. When one party to a conflict derives operational advantage from AI-assisted targeting or autonomous munitions, every opposing party faces pressure to match that capability or accept a structural disadvantage. Arms-race dynamics in conventional weapons typically proceed over decades. The AI-enabled systems now entering operational use have deployed, been evaluated under fire, and been refined on a timeline measured in months. The diffusion of this technology — across state actors, non-state actors, and the commercial sector that supplies both — is not a projection. It is an observation of what has already occurred.

The Structural Dimension

Vance's framing — that AI will change warfare — is accurate, but the change is not merely operational. It is structural. Autonomous systems, once deployed at scale, alter the relationship between force application and political authorisation in ways that do not require any single policy decision to initiate.

Consider the decision cycle in traditional warfare. A commander identifies a target, submits it for political or legal review if required, receives authorisation, and then executes. The bottleneck is human judgment, and human judgment is slow, subject to political constraint, and politically visible. Autonomous systems that can operate within a defined envelope — a geometric area, a set of electronic signatures, a behavioral pattern — have the effect of disaggregating that bottleneck. Force application becomes, in a meaningful sense, devolved from political to tactical level. The targeting decision migrates from the command authority to the algorithmic parameters set at configuration time.

This is not a critique of any particular government or military doctrine. It is an observation about what autonomous targeting architectures do by design. They encode political and legal judgments about proportionality, distinction, and military necessity into software parameters that are then executed at machine speed, without live human review of individual engagements. Whether that is desirable depends on whether one believes current human review processes are too slow or too constrained — and that question has different answers in different institutional contexts.

The competitive dimension reinforces this structural logic. If the United States does not field AI-assisted targeting systems in a given domain, adversaries who do gain a measurable advantage in that domain. This creates systemic pressure toward deployment regardless of what any single state believes about the wisdom of doing so. Arms control agreements governing autonomous weapons have been discussed at the Convention on Certain Conventional Weapons under the UN framework since at least 2013. Progress has been incremental. The technology has not waited.

China's civil-military fusion programme, which mandates that commercial AI research contributes to military capability development, positions it to move rapidly from laboratory to operational system in a way that is structurally different from the procurement pathways of Western democracies, where commercial AI development operates largely independently of defence acquisition. This is not a uniquely Chinese model — the US Defense Advanced Research Projects Agency has long pursued a similar integration function — but the institutional architecture differs in ways that affect pace and scale of deployment. These structural differences are now being studied intently by defence planners across NATO and in the Indo-Pacific.

The Governance Vacuum

International humanitarian law — the body of treaty and custom governing conduct in armed conflict — was designed around the assumption that lethal force is applied by human decision-makers who can be held accountable for violations. The principles of distinction (combatants versus civilians), proportionality (civilian harm versus military advantage), and precaution (feasible steps to avoid civilian harm) all presume a human agent capable of applying judgment, weighing evidence, and making choices that can be evaluated after the fact.

An autonomous system that selects and engages targets operates on none of these presumptions. It applies an algorithm trained on historical data to a specific engagement. The question of whether that algorithm correctly distinguished a civilian from a combatant, or correctly assessed proportionality, is not a question that can be answered by examining the system post-hoc in the same way that a human commander's decision can be interrogated. The decision trail is in the training data and the model parameters, not in a chain of command that can be traced through orders and authorisations.

This creates a legal and accountability gap that the existing framework has not resolved. The International Committee of the Red Cross has published guidance arguing that human control over individual targeting decisions must be maintained — that autonomous systems may be lawful in some contexts but only if a human remains in the loop for critical decisions. Several states have endorsed the position that meaningful human control is a prerequisite for lawful autonomous weapons use. But there is no binding treaty, no enforcement mechanism, and no agreed technical standard for what constitutes sufficient human control. The discussion at the CCW has produced non-binding reports and political declarations without translating into treaty law.

The United States has taken the position that autonomous weapons systems are subject to existing international humanitarian law, and that current US policy requires human involvement in lethal decision-making. The precise contours of that requirement — whether it applies at the individual engagement level or at the system-authorisation level — are not publicly specified in full. Russia and China have not endorsed the meaningful human control framework and have opposed binding restrictions. The asymmetry between those three positions is not incidental. It reflects a genuine divergence in how states assess the strategic value of autonomous systems and their willingness to accept constraints on deployment.

What Comes Next

The trajectory is not fixed, but it is constrained by structural forces that are not easily redirected. The economics of AI-enabled systems — lower per-unit cost as commercial components diffuse, scalability of production, reduced reliance on trained human operators — create pressure for adoption across state and non-state actors. The competitive dynamic, once a leading military deploys AI-assisted or autonomous systems in a credible conflict, forces responses from peers and near-peers. The governance framework, to the extent it exists, operates at the speed of multilateral negotiation, which is measured in years, not months.

Vance's expression of worry about how AI will change warfare is a recognition that this process is underway. Whether it prompts a serious effort to establish binding constraints — or whether it remains a statement of concern without follow-through — is the more consequential question. History suggests that arms control agreements governing new categories of weapons are more achievable before a technology is deployed at scale and before its advocates have developed entrenched interests in its continued development. That window is not yet closed on AI-enabled autonomous weapons, but it is not indefinitely open.

The alternatives are not reassuring. A world in which lethal autonomous systems proliferate without agreed rules is a world in which the distinction between combatant and civilian — already under severe pressure in contemporary conflicts — is further degraded by systems that cannot be held accountable in the manner that existing international humanitarian law requires. The question is not whether that world is coming. It is whether the current moment still offers an opportunity to shape it.

Monexus notes that the Vance quote was carried by Disclose.tv on the day it was delivered. Much of the substantive AI military development it gestures toward — drone proliferation, autonomous systems in Ukraine, UN governance discussions — is documented across the open-source defence reporting community. This piece draws on that body of work. A future investigation will examine the specific procurement and deployment decisions of leading military powers in greater detail.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://twitter.com/disclosetv/status/2060032986058227912/video/1twee
• https://t.me/osintlive
• https://t.me/disclosetv
• https://x.com/ekonomat_pl/status/2059988640428662784
• https://en.wikipedia.org/wiki/Lethal_autonomous_weapons_system
• https://en.wikipedia.org/wiki/2020_Nagorno-Karabakh_war
• https://en.wikipedia.org/wiki/UN_Convention_on_Certain_Conventional_Weapons
• https://en.wikipedia.org/wiki/Civil%E2%80%93military_fusion#Artificial_intelligence
• https://en.wikipedia.org/wiki/Future_of_Life_Institute