Law report No. GLW-2495 · filed November 17, 2026
LegislationReported case
Geneva talks face pivotal choice on autonomous weapons
Helen Clark says AI systems assisted with targeting 37,000 Palestinians in Gaza and urges Geneva states to negotiate binding limits on autonomous weapons.
By Marcus Bennett5 min read956 words
Holding
- Lavender assisted in the targeting of 37,000 Palestinians in Gaza after 7 October 2023, Helen Clark says.
- The United States used AI-enabled targeting to strike 1,000 targets in the first 24 hours of its February attacks in Iran, according to Clark.
- More than 120 children were killed when a school in Minab was bombed during those attacks, according to Clark.
- The seventh Convention on Certain Conventional Weapons review conference meets in Geneva this week, according to Clark.
- The New Zealand Nuclear Free Zone, Disarmament, and Arms Control Act 1987 bars nuclear weapons and nuclear-powered vessels from entering the country.

By Marcus Bennett
AI systems assisted with the targeting of 37,000 Palestinians in Gaza after 7 October 2023, Helen Clark says. The former New Zealand prime minister and former head of the United Nations Development Programme writes that states face a choice between launching treaty negotiations on autonomous weapons and allowing more such systems onto battlefields. The seventh review conference of the Convention on Certain Conventional Weapons meets in Geneva this week; the article gives no calendar date.
The conference is the decision-making body Clark urges to act. Her article presents a proposed negotiating mandate, not a completed legal rule. A binding treaty would govern the development and use of autonomous weapons systems.
Clark is also a member of The Elders.
How has AI changed the pace of war?
Militaries worldwide increasingly use AI to find, track and attack targets, Clark says. A strike that once required about half an hour and substantial human consideration can now take minutes or even seconds. In that model, an AI system takes the lead once it passes a confidence threshold set by a human operator.
The United States relied on AI-enabled targeting to strike 1,000 targets during the first 24 hours of its February attacks in Iran. A school in Minab was bombed during those attacks, killing more than 120 children, according to Clark.
In the Russia-Ukraine war, AI-enabled drones have received little or no human input during the final stage of a strike. Across the Middle East, Israel Defense Forces systems named “Lavender,” “The Gospel” and “Where’s Daddy?” have identified alleged militants and their homes as potential bombing targets.
Lavender assisted with the targeting of 37,000 Palestinians in Gaza in the months after 7 October 2023. A senior Israeli commander described the system as bypassing “a human bottleneck for both locating the new targets and decision-making to approve the targets.”
Clark calls these uses evidence of a wider global trend, rather than isolated cases. Her concern is legal as much as operational. Decisions about force must remain answerable to people and states, not machines or AI systems.
Why does legal accountability become harder?
Clark says anti-personnel weapons create the gravest risk because they target people rather than vehicles or buildings. They struggle to distinguish a combatant from a journalist, civilian or child. If an autonomous system assists in committing a war crime, determining who bears responsibility becomes more difficult.
Pope Leo XIV made the same moral point in the recent encyclical Magnifica Humanitas: “Any technology that facilitates attacks without seeing the face of human beings lowers the moral threshold of conflict.”
The technology has advanced faster than law. Clark says governments possessing the most advanced autonomous systems have resisted binding limits. Their hesitation has not halted weapons, data-processing systems or related surveillance tools. The distance between technical capability and regulatory restraint continues to widen.
Cost and replication add another risk. Some autonomous technologies are cheap to produce and easy to copy. Terrorist groups and other non-state actors could acquire them. Clark says every state has an interest in preventing that outcome.
Can states build a global consensus?
Clark points to a domestic precedent from nearly 40 years earlier. She played a role in securing passage of the New Zealand Nuclear Free Zone, Disarmament, and Arms Control Act 1987. The law bars nuclear weapons and nuclear-powered vessels from entering New Zealand. Clark says it has retained consistently high public support since its introduction.
She compares that achievement with the present need for comparable global ambition. The conference comes amid fractured international relations. Even agreement to negotiate would be a positive step. Clark believes sustained talks could produce binding agreements comparable to those that banned chemical weapons and anti-personnel landmines.
Her proposed endpoint is specific rather than a general promise. A meaningful treaty would establish human control, accountability and non-proliferation as priorities. Countries benefiting from the current accountability gap may resist, Clark warns. She urges the majority of states to act together and reject a weak settlement.
What should the treaty require?
Clark proposes two tiers of control:
- A ban on autonomous weapons systems that cannot comply with international humanitarian law.
- Regulation of every other autonomous weapon through limits on engagement and targeting, human supervision, and testing and evaluation.
The proposal is not an adopted rule. No conference decision or treaty text appears in Clark’s account. Governments would still need to define which systems fall within the ban and how supervision, testing and accountability would work in practice.
The broader warning concerns timing. Technology will continue developing while diplomats seek consensus. If states do not seize the current opportunity, Clark says, collective security faces greater danger.
What should legal advisers watch?\n
For government counsel and military advisers, the practical consequence is a defined drafting agenda. Human control, accountability and non-proliferation would need rules on engagement and targeting, supervision, testing and evaluation. Drafters would also define which systems cannot comply with international humanitarian law and fall within the proposed ban. The controls would remain proposals until adopted.
State delegations would then face the unresolved accountability question when an autonomous system assists in committing a war crime. Clark places that problem at the center of the proposed regime, rather than treating machine decision-making as a substitute for state responsibility.
Clark’s own example supplies the closing argument. New Zealand’s 1987 act shows that ambitious law can endure when states act together. Geneva now faces a similar choice on a faster-moving technology. The conference can launch negotiations, or permit more autonomous systems to reach battlefields, Clark says. A weak treaty, in her view, would reward the governments least willing to accept accountability.
via theguardian.com (Source)
More from Marcus Bennett
Show full bio
Market editor covering regulation, policy and their impact on International law and legal practice at Global Law Wire.
319 articles
Also before the court
- Geneva Talks on 'Killer Robots' Meet Five-Year Clock and Great-Power Resistance
- CCW GGE Adopts 'Control and Human Judgement' Formula for LAWS Report
- CCW GGE Adopts Consensus Characterisation of Lethal Autonomous Weapons
- Autonomous Weapons Kill Civilians. Victims Get No Reparations Framework
- Global Courts Year-End 2025: Rulings That Reshaped Legal Frontiers