🤖 Robonomy

Autonomous Weapons Governance — Where the Law Actually Stands

6 min read

The Robots are making targeting decisions on the battlefield. The law has not caught up.

In April 2026, a ukraine-25000-ugvs-frontline-logistics. combat-ground-robot-fleet-statewatch|Russia is running 32 different models of combat ground robots]], largest producers unsanctioned. scout-ai-100m-series-a-fury-unmanned-warfare-brain to build an AI brain that manages autonomous strike missions.

The weapons are ahead of the law. By a lot.

Here’s what the law actually says, where it breaks down, and what that means for every autonomous robot in this database.

What International Humanitarian Law Actually Requires

International Humanitarian Law (IHL) — the laws of war — was built for humans. Its core principles:

Distinction: Parties must distinguish between combatants and civilians. Attacking civilians is prohibited.

Proportionality: An attack that causes excessive civilian harm relative to military advantage is prohibited.

Precaution: Parties must take feasible precautions to minimize civilian harm.

Here’s the problem: all three require judgment. Is that person a civilian? Is this attack proportional given current battlefield conditions? Is the precaution feasible? These are contextual, probabilistic, human decisions.

An autonomous weapons system making a kill decision faces the same requirements. IHL doesn’t care whether the thing pulling the trigger is a person or a machine. The obligation is the same.

The question isn’t “can robots comply with IHL?” The question is “whose fault is it when they don’t?”

The Accountability Gap

When a human soldier violates IHL, the legal chain is relatively clear: the soldier, their commanding officer, potentially the state. Not clean, but structured.

When an autonomous weapons system violates IHL — targets a hospital, kills civilians, fires on the wrong person — accountability evaporates. The programmer? The commander who deployed it? The manufacturer? The state?

Current IHL doesn’t have a good answer. This is the “accountability gap” that legal scholars and arms control advocates have been raising since at least 2013. Over a decade later, it’s still a gap.

What has happened:

  • No binding treaty on autonomous weapons exists
  • No international body has authority to regulate them
  • No state has banned fully autonomous weapons domestically
  • Several states actively developing autonomous targeting systems

The CCW Process: Twelve Years of Deliberation

The main international forum is the Convention on Certain Conventional Weapons (CCW) — the same body that produced treaties on land mines and cluster munitions. Since 2014, states have been meeting under a Group of Governmental Experts (GGE) to discuss “LAWS systems” (LAWS).

Twelve years of discussion. Zero binding outcome.

The 2023 CCW GGE sessions produced “guiding principles” — not rules, not a treaty, not enforcement mechanisms. The principles include things like:

“Human responsibility for decisions on the use of lethal force must be retained.”

That sounds good. But it’s not legally binding. And it doesn’t define what “human responsibility” means when a system makes autonomous targeting decisions 40 times per second.

The CCW requires consensus, which means any state with strategic interest in autonomous weapons can block progress. Russia, the United States, Israel, and China have all, at various points, stalled or undermined binding commitments.

The US Position

The US Department of Defense issued Directive 3000.09 in 2012, revised in 2023. It requires that autonomous weapons systems be designed to allow “appropriate levels of human judgment over the use of force.”

“Appropriate” is doing enormous work there. The directive doesn’t prohibit autonomous targeting — it requires human judgment to be in the loop at “appropriate” levels. Military lawyers have been parsing what that means ever since.

The US position at the CCW has generally been: existing IHL is sufficient, states just need to ensure their autonomous systems comply with it. In other words: no new rules needed, just apply the old ones correctly.

This position conveniently avoids binding constraints while maintaining the US’s freedom to develop autonomous systems. [UNCERTAIN — US position has evolved; check latest CCW GGE documentation for current stance]

The AI Acceleration Problem

The governance debate was already struggling when the pace of AI development was gradual. Now it’s exponential.

In 2024-2026 alone:

  • Scout AI’s Fury model demonstrated autonomous end-to-end strike mission execution
  • Ukraine deployed 25,000 UGVs with AI-assisted targeting
  • Russia fielded kamikaze drones with autonomous terminal guidance
  • Multiple companies built commercial VLA Model models that could be adapted for weapons control in months

Legal processes move in years. AI development moves in months. The gap is widening.

The GGE experts meeting every six months to produce non-binding guidance on systems that didn’t exist when they last met is theater. It signals concern without producing change.

What’s Actually Happening Instead

Since the multilateral process has stalled, states are moving bilaterally and unilaterally:

The Responsible AI in the Military Domain (REAIM) summit (2023): Produced a “call to action” signed by 60+ countries. Not binding. No enforcement mechanism.

ICRC position: The International Committee of the Red Cross has repeatedly called for a legally binding treaty prohibiting autonomous weapons that select and attack humans without meaningful human control. States haven’t followed.

Commercial technology transfer: The dual-use problem. Every commercial autonomous system — a Spot quadruped, a warehouse drone, a navigation AI — can be adapted for military use. Export controls are patchwork. The Scout AI investors include defense contractors; the line between commercial and military capability is already blurred.

My Read

The governance situation is not a temporary policy gap waiting to be filled. It’s a structural failure of the international system to regulate a technology that genuinely challenges existing legal frameworks.

Here’s the thing: IHL works when accountability is traceable. When the thing that killed someone is a machine, and the machine was built by a company, purchased by a state, and deployed by a commander — accountability doesn’t dissolve, but it distributes so widely that enforcement becomes practically impossible.

What I’d actually watch:

  1. The first international incident with clear autonomous targeting violation. A documented case of an AI making a targeting error resulting in civilian deaths, with evidence of autonomous decision-making, would force the legal debate from theoretical to operational.

  2. The accountability case. Which state will be the first to face genuine IHL accountability for autonomous weapons violations? How that case resolves — or fails to resolve — will define the next decade of governance.

  3. The hardware/software split. Current governance frameworks focus on the weapon. They should focus on the AI model. Regulating a robot is tractable. Regulating a weights file that can be deployed in any vehicle is not.

The robots aren’t waiting for the law. The law needs to move faster.

Sources

  • CCW GGE session records 2014–2023
  • DoD Directive 3000.09 (2023 revision)
  • ICRC reports on autonomous weapons
  • REAIM summit documentation

Last updated: May 2026 | Verification: partially-verified | Status: Ongoing international deliberation, no binding outcome

Graph View

Backlinks

Recent