The three words that will decide whether robots can kill people in war

Key Points:

• A Russian AI-enabled drone in Ukraine autonomously selected its target, resulting in the deaths of three civilians, marking a significant and troubling instance of lethal machine decision-making without direct human input.
• The international debate on autonomous weapons centers on two frameworks: “meaningful human control,” which demands human involvement at the moment of attack, and the US-led “appropriate human judgment,” which emphasizes human oversight throughout a weapon’s lifecycle rather than at the point of firing.
• The US framework allows flexibility in human involvement based on context, shifting from strict control to ensuring autonomous systems adhere to legal and ethical standards, but it leaves open the critical question of who determines what level of human judgment is “appropriate.”
• OpenAI recently adopted the Pentagon’s “appropriate human judgment” language in its military use policy, signaling a willingness to accept ambiguous standards of human control over lethal AI applications, which raises concerns about transparency and accountability.
• As autonomous weapons technology advances rapidly with limited international safeguards, clarifying the meaning and limits of human oversight is urgent to prevent losing control over machine-driven lethal decisions in warfare.