No International Consensus on When AI Can Autonomously Decide to Kill

Autonomous weapon systems that can select and engage targets without human intervention are already being deployed, yet there is no binding international law governing their use. Russia deploys 30-50 autonomous strike drones daily in Ukraine. The V2U drone, first observed in 2024, autonomously navigates GPS-denied environments and identifies targets. Multiple nations are developing systems where the decision to kill is delegated partly or fully to algorithms, and U.S. policy explicitly does not prohibit lethal autonomous weapons. This matters because without a legal framework, there is no enforceable standard for when a machine may take a human life. Human Rights Watch documented that autonomous systems lack the contextual awareness to reliably distinguish between combatants and civilians, with people using wheelchairs particularly at risk because their mobility aids can be misidentified as weapons by computer vision systems. When a drone kills the wrong person, existing international humanitarian law cannot clearly assign criminal responsibility: the programmer, the commander who deployed it, the procurement officer who approved it, or the algorithm itself. The arms race dynamic makes restraint irrational for any single nation. Senior U.S. military leaders have stated that the United States may be compelled to develop autonomous weapons if competitors do so. China, Russia, Israel, Turkey, and Iran are all advancing autonomous capabilities. The UN Secretary-General has called for banning machines with fully delegated lethal authority, but the nations building these systems have blocked binding resolutions. The structural reason this persists is the same dynamic that delayed nuclear arms control: the technology provides such decisive military advantage that no major power will voluntarily constrain itself before its rivals do. But unlike nuclear weapons, autonomous drones are cheap, proliferating rapidly to non-state actors, and do not require massive industrial infrastructure. By the time a treaty is negotiated, the technology may be too widespread to regulate.