In early September, hundreds of diplomats gathered in the UN conference hall in Switzerland to agree on an international document intended to establish rules for the use of lethal autonomous weapons. This marks the first attempt to create an agreement regulating the use of weapons based on artificial intelligence (AI). However, by the final day, the negotiations had reached a tense phase. According to three sources familiar with the discussions, as well as documents, American and Russian diplomats spent approximately 15 hours actively editing the text of the agreement. As a result, several provisions providing for additional safety guarantees regarding the use of AI in the military sphere were removed. This was reported by Dengi.ua, citing The Washington Post.
Specifically, requirements for the “predictable” and “reliable” operation of such systems were removed from the draft. A provision mandating consideration of ethical aspects when using weapons created with artificial intelligence was also omitted. Furthermore, U.S. and Russian experts removed a provision requiring a human to verify military targets identified by AI before a strike could be carried out.
According to sources, the most intense work took place during a closed-door session, during which UN cameras were turned off and civil society representatives were removed from the room. The U.S. and Russia each sent about 10 lawyers to the negotiations - nearly twice as many as the legal representation of most other delegations. Changes were made so rapidly that it was difficult for smaller delegations to keep up.
The U.S. State Department, the Russian Ministry of Foreign Affairs, and the UN did not respond to the publication’s requests for comment.
The watering down of the agreement’s provisions illustrates how Washington and Moscow opposed stricter international restrictions on lethal AI-based systems. At the same time, the Pentagon is seeking to accelerate the integration of rapidly advancing artificial intelligence technologies into its own military strategies.
The negotiations are taking place against the backdrop of a wide-ranging debate on the potential risks of AI development. Technology executives are calling on nations to jointly establish rules and restrictions for this technology. The UN has been trying to develop norms for the military use of artificial intelligence for several years, but has failed to make significant progress. The current directive is the most notable result of these efforts.
The U.S. actions during the negotiations may also provide insight into Washington’s approach to regulating AI within its own armed forces. The U.S. government is currently reviewing its existing directive on lethal autonomous weapons, adopted in 2023.
Verity Coyle, deputy director for arms issues in the Crisis, Conflict, and Arms Division at Human Rights Watch, warned that insufficiently strict international norms in this area could lead to serious consequences.
“This could mean that machines will be able to make life-and-death decisions without human involvement. The practical consequences are clear: more harm to civilians, less accountability, and a faster slide toward riskier, automated warfare,” Coyle said.
Meanwhile, the use of AI in armed conflicts is already moving beyond theoretical scenarios. According to the publication, the U.S. military used the Anthropic chatbot to identify nearly 1,000 targets for strikes during the first day of a war with Iran. Israel, as reported by the publication, also made extensive use of artificial intelligence during military operations in the Gaza Strip.
Further concerns are raised by the expansion of AI capabilities beyond the immediate military sphere. In recent months, software developers have reported instances where their systems independently hacked into the servers of other organizations, including U.S. government websites. In addition, malicious actors in countries such as Yemen and Iran have used chatbots to attempt to develop weapons or plan attacks against U.S. military personnel. This heightens concerns about how controllable the use of such technologies will remain in the future.
