ICRC legal expert warns AI targeting is outpacing meaningful human control

AI is speeding up military targeting while putting meaningful human control under strain, Laurent Gisel, head of the International Committee of the Red Cross’s legal warfare team, warned in an interview published by 404 Media on October 22. He described the technology as already amplifying destruction and humanitarian harm—not merely posing a future threat.

Gisel’s warning concerns both weapons that select and strike targets themselves and AI tools that help people make targeting decisions. His proposed response is not a blanket ban on military automation. He outlined prohibitions for specific autonomous weapons, alongside restrictions intended to preserve human judgment over other uses.

Faster decisions, less room to question them

The danger, in Gisel’s account, is not speed alone. Targets are being generated more frequently and attacked more quickly, raising questions about how much control people actually retain. He warned that humans tend to trust automated systems, especially when tired, stressed or under pressure to deliver quickly—the conditions soldiers face.

He also cautioned that developers may be far removed from the battlefield. Some systems come from technology companies whose products were not originally designed for combat, he said. That distance can leave developers unaware of the consequences when their tools help guide the use of force.

If you have a weapon that is unpredictable in its effect, how can you deliver the commander's intent?

Laurent Gisel, speaking to 404 Media

The dividing line is who selects and strikes

The ICRC’s October 8 explainer makes a distinction: autonomous weapons are systems that, once activated, select targets and apply force without further human intervention. Sensors match objects or people against a predefined target profile, and the weapon can initiate a strike without a person choosing that specific target.

That does not describe every armed drone or every military use of AI. A remotely piloted drone whose human operator identifies the target and decides when to fire is not autonomous under this definition. AI used only for navigation does not necessarily make a weapon autonomous, either. Conversely, autonomous weapons can use simple rules rather than AI.

The organization’s existing position calls for new, legally binding international rules with three main elements:

• Prohibit autonomous weapons whose effects cannot be sufficiently understood or predicted.
• Prohibit autonomous weapons designed to select and attack human beings directly.
• Place strict restrictions on all other autonomous weapons.

Accuracy would not settle the moral question

For weapons that target people, the ICRC’s objection extends beyond technical reliability. Its explainer argues that a person’s status can change rapidly: someone may surrender, become wounded or cease participating in hostilities. Those judgments depend on conduct and context, not simply a match between sensor readings and a target profile.

The organization also draws a moral line against delegating life-and-death decisions about people to machines, regardless of how accurate the technology becomes. Existing international humanitarian law already applies to autonomous weapons, but the ICRC argues that its requirements cannot simply be coded into a system in advance.

Gisel described resistance even to safeguards that might seem aligned with military discipline. Recalling negotiations in August and September, he said language requiring autonomous weapons to be reliable, explainable, understandable and predictable had been removed from a proposal by some states. His interview did not present those safeguards as newly agreed rules.