AI Safety

Autonomous Weapons Governance: What an International Treaty Would Need to Do

Lethal autonomous weapons are being developed and deployed without international governance. What would a credible binding treaty need to say, and is it politically achievable?

← Back to Top 10 Existential AI Threats

Lethal autonomous weapons systems (LAWS) are weapons that can select and engage targets without meaningful human control. They represent a qualitative shift in warfare: the removal of human judgment from the kill decision. As AI capabilities advance, autonomous weapon systems are becoming more capable, cheaper, and more widely proliferated. The governance gap is real and growing.

What is meaningful human control?

The central concept in autonomous weapons governance is "meaningful human control." Most advocates for LAWS restrictions argue that humans must retain meaningful control over targeting decisions because human life cannot ethically be taken by an algorithm without human responsibility and judgment. International humanitarian law (IHL) requires that attacks be assessed for proportionality and distinction between combatants and civilians. These assessments require contextual judgment that current AI systems cannot reliably make.

Meaningful human control means more than a human somewhere in the loop: a human pressing a button to launch a pre-programmed attack sequence is not exercising meaningful control over the target selection. It requires that the human understands the specific target, can assess proportionality, and retains the ability to withhold or cancel the attack based on that assessment.

The governance landscape as of 2026

Discussions on LAWS governance have been ongoing at the UN Convention on Certain Conventional Weapons (CCW) since 2014. Progress has been slow. A growing coalition of states, led by Austria, New Zealand, and supported by the ICRC, supports negotiating a binding treaty. But major military powers including the US, Russia, and China have resisted binding obligations, preferring non-binding guidelines and voluntary political declarations.

Outside the CCW framework, several bilateral and multilateral initiatives are relevant. NATO allies have agreed political commitments to meaningful human control in alliance operations. The Political Declaration on Responsible Military Use of Artificial Intelligence (2023) attracted over 50 state signatories. But political declarations are not legally binding and create no enforcement mechanisms.

What a credible treaty would need

For an international treaty on LAWS to be effective, it would need to address at least five elements:

The proliferation challenge

Even a strong treaty faces a proliferation challenge that does not apply to, say, chemical weapons: autonomous weapon software is dual-use, easily copied, and can be built by non-state actors with sufficient technical capability. A treaty regime would need to include non-state actors and address the asymmetry between the cost of deploying an autonomous weapon (potentially very low) and the cost of defending against it.

This does not make governance impossible, but it does mean that a treaty alone is insufficient. Complementary measures including export controls on enabling AI components, integration of LAWS governance into the broader AI governance agenda, and investment in international monitoring and response capacity are all necessary parts of a serious policy response.

AI safety governance is one of the core topics in the Better Societies community. Join the conversation.

Related reading

Frequently asked questions

What are lethal autonomous weapons systems?

LAWS are weapons that can select and engage targets without meaningful human control, using AI to identify, track, and decide to attack based on pre-set criteria. The defining characteristic is the removal of direct human decision-making from the engagement sequence.

Is there an international treaty on autonomous weapons?

As of mid-2026, no binding international treaty exists. Discussions have been ongoing at the UN Convention on Certain Conventional Weapons since 2014. A growing coalition supports a binding treaty, but major military powers including the US, Russia, and China have resisted binding obligations.

What is meaningful human control in weapons?

Meaningful human control means a human operator retains sufficient understanding and judgment to be responsible for lethal force decisions. It includes reviewing the AI targeting recommendation, understanding context, and being able to withhold or cancel an attack. It is distinct from a human simply switching the system on.

Join the AI safety community

Connect with researchers, policymakers, and practitioners working on AI governance. Free to join.