{
  "id": 9173466,
  "title": "Soldiers can refuse to commit war crimes. Can AI?",
  "url": "https://urgent.news/2026/09/22/soldiers-can-refuse-to-commit-war-crimes-can-ai",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-22T18:10:00.000Z",
  "source": {
    "name": "Vox",
    "slug": "vox",
    "url": "https://www.vox.com/future-perfect/503564/military-ai-minimally-refuse-war-crimes"
  },
  "original_language": "en",
  "account": "The Pentagon's push to loosen limits on how AI companies may use their technology in war has raised a significant question: who gets to say \"no\" to a machine? When Anthropic refused to bend to the Pentagon's more flexible standard, it lost a major government deal. OpenAI, however, embraced the Pentagon's demand for AI models that rarely \"refuse\" commands. The agreement included specific language specifying the type of AI models the Department of Defense (DOD) would receive, emphasizing models with minimal refusal rates.\n\nOn the surface, this provision might seem reasonable, given the strictly hierarchical nature of the military. Soldiers are generally expected to follow their commanders' orders without question, especially in high-stress and time-sensitive situations. However, the military is not the only institution that requires obedience. Soldiers are legally required to follow lawful orders and can refuse unlawful ones. This principle is rooted in US domestic, military, and international law, as well as established common law principles.\n\nThe Department of Defense's Law of War Manual clearly outlines that each member of the armed forces has a duty to comply with the law of war in good faith and to refuse to comply with clearly illegal orders. The Nuremberg Trials, which dealt with Nazi soldiers after WWII, rejected the defense of obedience without limits, emphasizing that individual moral and legal responsibility supersedes military chains of command for illegal acts.\n\nBut what about machines? When using an AI chatbot like ChatGPT, the system does not reject requests it disagrees with or likes. Instead, it follows established protocols, refusing requests that violate its terms of service. The military's use of AI for more consequential applications, such as autonomous weapons, raises the question of whether the same logic applies to agentic AI systems.\n\nRebecca Crootof, a technology law expert, noted that while the Pentagon's language could increase compliance with unlawful orders or reduce friction around them, it is not a guarantee of compliance with unlawful orders. Identifying and evaluating what constitutes an unlawful order in the moment can be incredibly difficult, even for human soldiers. The fact that soldiers can disobey orders is an important constraint on commanders who might let battlefield conditions or superior orders erode their ethics.\n\nAs AI systems become more prevalent in warfare, the question of whether they should be allowed to refuse commands becomes increasingly relevant. The Pentagon's request for AI with minimal refusal rates may create a world where machines can act with less accountability than human soldiers, potentially increasing the likelihood of war crimes.",
  "summary": "Who gets to tell the military “no”? That question has become even more important after the Pentagon pushed AI companies to loosen limits on how their technology may be used in war — including how much human control must remain over autonomous weapons. When Anthropic said no, it lost a major government deal. OpenAI stepped […]",
  "key_points": [
    "Soldiers can refuse unlawful orders in US law.",
    "AI chatbots refuse requests violating terms of service.",
    "Pentagon's AI demand may reduce accountability for war crimes."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}