Many experts view 2026 as the last realistic window to agree on rules for autonomous killing machines. Current geopolitical trends suggest a binding agreement remains unlikely.
The Situation in Five Lines
- The UN Secretary-General set a 2026 deadline for a legally binding treaty on lethal autonomous weapons systems (LAWS).
- 156 nations have backed regulation. The US, Russia, and China have refused to be bound by it.
- At the 2026 REAIM Summit, both the US and China declined to sign the declaration on responsible military AI.
- Ukraine's war with Russia has turned autonomous drones into the defining weapon of modern conflict, accelerating deployment far faster than governance can follow.
- Many analysts describe 2026 as the last realistic window for any treaty before proliferation makes regulation practically meaningless.
Key Terms
| Term | What It Means |
|---|
| LAWS | Lethal Autonomous Weapons Systems: weapons that can select and engage targets without a human making the final firing decision. |
| CCW | Convention on Certain Conventional Weapons: a UN-linked arms control treaty forum where LAWS discussions have taken place since 2014, without binding result. |
| REAIM | Responsible AI in the Military Domain: a multi-nation summit series aimed at building consensus on governing military AI. Previous summits were held in The Hague (2023) and Seoul (2024). |
| Replicator Programme | A US Department of Defense initiative to rapidly deploy thousands of low-cost autonomous drones and surface vessels across multiple military domains. |
| IHL / Humanitarian Law | International Humanitarian Law: the rules of war that require combatants to distinguish between military targets and civilians. The central legal question is whether an autonomous system can reliably do this. |
| Pre-Proliferation Window | The period, believed by many analysts to be closing now, before autonomous weapons spread too widely for any treaty to meaningfully constrain them. |
Timeline: Twelve Years of No Agreement
| Year | What Happened |
|---|
| 2014 | CCW Group of Governmental Experts begins formal discussions on lethal autonomous weapons in Geneva. |
| 2018 | UN Secretary-General Guterres calls LAWS "politically unacceptable and morally repugnant." Calls begin for prohibition. |
| 2023 | Austria introduces the first UN General Assembly resolution on LAWS, backed by 164 states. Guterres sets a 2026 treaty deadline in his New Agenda for Peace. |
| Dec 2024 | UNGA adopts Resolution 79/239 with 166 votes in favour. Russia votes against; US abstains. |
| May 2025 | First UN General Assembly meeting on autonomous weapons attended by 96 countries. No binding outcome. |
| Oct 2025 | US reverses its support for a UN resolution on responsible military AI, citing sovereignty concerns. |
| Nov 2025 | 156 nations support a UNGA resolution calling for a binding LAWS agreement by the 2026 CCW Review Conference. |
| Feb 2026 | REAIM Summit in A Coruña: only 35 of 85 countries sign the declaration. US and China decline. |
| 2026 | Seventh CCW Review Conference: the stated deadline for a legally binding instrument. Outcome uncertain. |
| 2028–2030 | Window analysts describe as the point after which autonomous weapons proliferation may become irreversible. |
Where We Are Right Now
Since 2014, nations have been gathering in Geneva under the Convention on Certain Conventional Weapons to discuss lethal autonomous weapons. Twelve years of meetings. No binding outcome. The group produces what diplomats call a "rolling text," a document of potential regulatory measures that accumulates language without producing law.
In December 2024, 166 nations voted in the UN General Assembly for a resolution calling for negotiations on an enforceable agreement by the Seventh Review Conference in 2026. Only three voted against: Belarus, North Korea, and Russia. The United States abstained. That vote, celebrated by many as a breakthrough, actually revealed the structural problem clearly. The countries with the most advanced autonomous weapons programs were either against or absent.
Then came February 2026. At the REAIM Summit in A Coruña, only 35 of 85 participating countries signed the declaration on responsible military AI principles. The United States and China, the two dominant military AI powers on the planet, declined. The Dutch Defence Minister, in what was perhaps the most honest statement made at the entire summit, described it as a global prisoner's dilemma where no country wants to unilaterally limit itself against adversaries who will not do the same.
What the Technology Actually Looks Like
The debate around "killer robots" often conjures science fiction imagery that obscures how grounded the actual threat already is. In January 2026, China's PLA National University of Defence Technology broadcast footage of a single soldier controlling a swarm formation of 200 autonomous drones. The Pentagon, according to multiple reports, is concerned it cannot match China's manufacturing pace in this domain.
In Ukraine, autonomous drones have ceased to be a novelty. By March 2026, they reportedly accounted for 96 percent of Russia's 35,551 battlefield casualties in that month, according to Ukrainian Defence Ministry figures. Ukrainian drones killed or seriously injured more than 240,000 Russian soldiers across 2025, the ministry claimed. These numbers come from a wartime government and cannot be independently verified; Russian figures differ significantly. What is not in dispute is the scale and direction of the shift: autonomous drone deployment has reorganised an entire war effort, and the change is irreversible.
The US is not standing still either. The Pentagon's Replicator programme reportedly received one billion dollars in 2025 to fast-track the deployment of thousands of expendable autonomous drones and surface vessels. For fiscal year 2026, Pentagon budget request documents indicated $14.2 billion allocated for AI and autonomous systems research. Israel, meanwhile, has accelerated deployment of the Iron Beam, a laser system that uses autonomous targeting to neutralise incoming threats faster than any human operator could respond.
The technological reality makes the diplomatic conversation feel like it is happening at a different speed entirely.
Why the Major Powers Will Not Sign
The three powers whose signatures would make any treaty meaningful have each arrived at refusal through different reasoning, and understanding the distinction matters.
The United States argues that existing international humanitarian law, implemented through national doctrine, already covers autonomous weapons. Its Department of Defense has its own Ethical Principles for AI, which it considers sufficient. In October 2025, the US reversed its previous support for a UN resolution on responsible military AI, stating that such a regime risked creating centralized control over critical technology and violating national sovereignty. The current administration has signalled it will prioritize "innovation and operational flexibility" over international consensus.
Russia's position is simpler. It has flatly rejected any moratorium on developing autonomous weapons and has accused other nations of stalling tactics inside the CCW process while simultaneously blocking progress in that same forum. Russia's focus has been on deployable technology for the Ukraine front, not treaty frameworks.
China's calculation is the most interesting. It has been more ambiguous, at times supporting language around human oversight in weapons decisions while refusing to be bound by it in practice. China's industrial capacity to manufacture autonomous hardware at volume is itself a form of strategic leverage. Signing a treaty that restricts production would foreclose an advantage it has built.
The three positions are different, but they converge on the same outcome: no binding constraint.
Three Scenarios for the Next Decade
How to read these scenarios: These are grounded projections based on current trajectories, not predictions. The further the timeline, the wider the uncertainty. None of these is inevitable.
Best Case: A Partial Framework Holds
A treaty fails to materialise in 2026, but a coalition of willing states, perhaps 80 to 100 nations, establishes a binding protocol among themselves covering specific categories: systems that cannot identify the difference between combatants and civilians, systems that once activated cannot be deactivated, and weapons that select targets entirely without human review. Major powers do not join but tacitly observe some limits to maintain plausible diplomatic relationships. Non-proliferation becomes the focus, with export controls preventing the most dangerous systems from reaching non-state actors. This is feasible but requires political will that has not yet been demonstrated.
Most Likely: Fragmentation and Informal Norms
No binding treaty. Instead, the world sees a patchwork: bilateral understandings between specific nations, unilateral doctrines published by major powers claiming compliance with humanitarian law, and a loose set of norms that powerful states follow selectively. The CCW process continues without binding outcome. Smaller nations build limited autonomous capabilities. Non-state armed groups acquire progressively more advanced drone technology. By 2030, some form of autonomous weapon has been used in a way that triggers a significant civilian casualty event, producing a political shock that either accelerates or permanently kills the treaty effort depending on who is blamed.
Worst Case: The Pre-Proliferation Window Closes
The 2026 deadline passes without agreement. The Seventh Review Conference produces a statement of principles and nothing enforceable. By 2028 to 2030, autonomous drone swarms are standard military equipment for a dozen state actors and within reach of wealthy non-state groups. An incident involving civilian mass casualties attributed to an autonomous system occurs, but attribution itself is contested because autonomous systems complicate accountability chains. No state can be conclusively blamed. No international tribunal has jurisdiction. The incident produces calls for a treaty, but by that point the technology has diffused far enough that any treaty covers only a fraction of deployed systems. Regulation becomes permanently retrospective, always catching up to yesterday's technology.
What Treaty Failure Actually Means for People
Geopolitical discussions of autonomous weapons tend to stay in the language of strategic balance and state actors. That framing obscures who actually bears the cost when these systems proliferate without rules.
Between 2010 and 2025, the number of non-state armed groups deploying drones in attacks grew from 10 groups to 469 groups across 17 countries, according to ACLED data. Terrorist groups in Syria used explosive-laden drones to kill more than 89 soldiers and civilians in a single attack in 2023. The Revolutionary Armed Forces of Colombia and the National Liberation Army have adopted drone warfare, with 77 attacks recorded between 2024 and 2025 alone, according to the same dataset. This is the democratisation of aerial lethality. Autonomous targeting capability is the next step on that curve.
The ordinary person in a conflict zone faces a world where an AI system, with no human reviewing the decision at the moment of firing, may determine whether they live or die. If that system makes a wrong classification, there is no one to hold accountable in any meaningful way. The existing frameworks for war crimes require identifiable human decision-makers. Autonomous weapons, by design, remove that human from the loop at the critical moment.
The Prisoner's Dilemma Nobody Is Talking About Directly
| Actor |
Stated Position |
Actual Behavior |
Strategic Interest in No Treaty |
| United States |
Regulation via national doctrine |
Declined REAIM 2026 declaration; reversed support for UN AI resolution |
Preserve Replicator programme, AI software edge |
| China |
Human control in principle |
Declined REAIM 2026; expanding swarm capability rapidly |
Preserve manufacturing volume advantage |
| Russia |
Explicitly opposed to treaty |
Voted against; deploying autonomous systems in Ukraine |
Operational advantage on existing front lines |
| Israel |
Regulation impedes national security |
Voted against; Iron Beam autonomous targeting active |
Defensive autonomous systems are existential, not optional |
| EU/smaller states |
Support binding treaty |
Signed REAIM; pushing CCW rolling text |
Would accept treaty but cannot compel others |
The Dutch minister's "prisoner's dilemma" framing is accurate and worth sitting with. In game theory, a prisoner's dilemma produces the worst collective outcome when each actor rationally pursues individual self-interest. If you limit your autonomous weapons and your adversary does not, you have ceded a strategic advantage for nothing. So no one limits. Everyone loses, but only slowly, and the costs arrive diffused across civilian populations and future conflicts rather than immediately and visibly on the actors making the decision.
The only historical escape from a prisoner's dilemma in arms control has been when both sides faced a credible shared catastrophe large enough to override individual strategic calculation. Nuclear weapons produced that credibility because the mutual destruction was immediate, certain, and symmetric. Autonomous weapons have not yet produced a catastrophe of that scale. The question is whether the world waits for one.
Editorial: What I Think Is Really Happening
Editorial Analysis · Santhosh Vishwamitra
I have been watching this process for a while, and the thing that strikes me most is not the lack of a treaty. It is the quality of the refusal.
When the US reversed its support for the October 2025 UN AI resolution, it framed the objection around sovereignty and innovation. That framing is interesting because it is the same language used to resist chemical weapons controls, landmine bans, and cluster munitions treaties, all of which eventually produced binding agreements when the political cost of being outside them became high enough. The US is not outside the Chemical Weapons Convention. It eventually signed the cluster munitions protocol. The pattern suggests that the refusal today is a negotiating posture as much as a principled position.
But there is something structurally different this time. Previous arms control agreements dealt with weapons that required significant infrastructure to produce. You cannot manufacture nerve agents in a garage. You can manufacture a lethal autonomous drone with commercially available components, open-source AI models, and a small engineering team. The barrier to entry is dropping faster than the diplomatic process can move. That asymmetry is new.
What most people misread in this debate is the role of China. China's ambiguity is not a sign of undecidedness. It is a sign that China wants to keep a seat at whichever table eventually forms while preserving the freedom to build. It is running the same playbook it used in climate negotiations for years: support the goal in principle, negotiate the timelines and specifics endlessly, maintain operational freedom in the interim. That approach works as long as the international community is more interested in Chinese participation than in a binding result. Right now, it is.
The three things I think the next five years will reveal are these. First, whether a significant autonomous weapons incident, one with clear civilian casualties and unclear human accountability, produces the kind of political shock that landmine campaigns or chemical weapons use in Syria did. Second, whether smaller powers with genuine interest in a treaty, the Europeans, Australians, Canadians, and much of the Global South, are willing to form a binding coalition without the major powers and accept that such a treaty will cover only part of the world's arsenal. Third, whether export controls on the components and AI systems that make advanced autonomous targeting possible become a form of de facto regulation even without a formal treaty.
My honest read is that a binding global treaty before 2030 is unlikely. A partial coalition agreement among willing states, covering specific prohibited categories, is possible. And some form of incident in the next three to five years that reshapes the political conversation is, based on the trajectory of current deployments, close to inevitable.
The 2026 deadline the Secretary-General set is almost certainly going to pass without a binding instrument. That is not a failure of ambition. It is a reflection of who benefits from the absence of rules.
Key Takeaways
- 156 nations support regulation. The three countries whose participation would make a treaty effective have refused to be bound.
- Both the US and China declined to sign the REAIM 2026 declaration, the most concrete multilateral AI governance effort to date.
- Autonomous weapons are already operating at scale in Ukraine, Gaza, and beyond. The technology is not hypothetical.
- The "pre-proliferation window" is the period before these systems spread irreversibly. Most experts believe 2026 represents the closing of that window.
- Non-state groups are already using advanced drone weaponry. Autonomous targeting capability will follow the same proliferation curve.
- A partial coalition treaty among willing states is more likely than a universal global agreement.
- The accountability gap is the most serious structural problem: autonomous systems remove the identifiable human decision-maker from the moment of lethal force, which current war crimes law requires.
If you like our brand new Future Scenario series, Read More from the same series:
Future Scenario: Predicting the Post-Oil Middle East Transition
Future Scenario: What Happens When Quantum Computers Break Encryption
Santhosh Vishwamitra
Web Developer and Founder of Vishwamitra Chronicles, launched in February 2026.
Comments
Post a Comment