AI in War: Who Takes the Blame?

Bellandi Insight

AI in War: Who Claims Responsibility?

Autonomous weapons, drone swarms, AI-driven targeting — these technologies are reshaping conflict faster than law can catch up. When a machine decides to strike, where does accountability lie?

Recent Developments & Battlefield Examples

  • In the Ukraine-Russia conflict, autonomous drones (like Gogol-M and swarm systems) are increasingly deployed, some capable of identifying and striking targets with limited human oversight.
  • At the UN, nations regroup to regulate lethal autonomous weapons (LAWS). Despite growing urgency, major powers resist binding restrictions in favor of national guidelines.
  • The UN General Assembly passed a resolution Dec 2024 condemning fully autonomous weapons, proposing a two-tier approach: ban some systems outright, regulate others.
  • Ethics & technical risks: unpredictability, black-box behavior, misclassification, reward hacking — all present serious dangers if lethal systems act without human oversight.
  • Directive 3000.09 (U.S.) mandates “appropriate levels of human judgment” remain in autonomous systems that use lethal force.

War is becoming a laboratory. As technology races ahead, the gap between deployment and regulation widens dangerously.

Ethical & Legal Challenges

  • Absence of humans-in-the-loop: Fully autonomous systems may act without direct human approval in the moment. :contentReference[oaicite:5]{index=5}
  • Blurred accountability: Who is liable — developer, commander, state, or the algorithm itself?
  • Violation of international law: Proportionality, distinction between combatants & civilians, necessity — ethical rules are hard to embed in code. :contentReference[oaicite:6]{index=6}
  • Arms race escalation: Cheap drones + AI lower entry threshold. The tech may spread to non-state actors, increasing conflict risk. :contentReference[oaicite:7]{index=7}

Your Turn

Question: If machines autonomously decide to kill — even under human-set goals — who should be held responsible? And how do we stop mistakes from becoming disasters?