US and Russian diplomats have removed key requirements for safety, ethics and human control from a draft UN global agreement on the use of artificial intelligence weapons. This was reported by The Washington Post, citing three people familiar with the negotiations and documents.
The changes were made at the end of talks in Geneva in early September. Hundreds of diplomats participated in the talks, and the US and Russian delegations worked for about 15 hours in closed session. Both sides brought in about 10 lawyers – significantly more than the legal teams of the other delegations.
Which guarantees have been dropped
The requirement that autonomous weapons systems operate in a “predictable“ and “reliable“ manner has been removed from the text. A clause stating that ethical considerations must be taken into account when using weapons with artificial intelligence has also been dropped.
The American and Russian representatives have also deleted a provision that required a person to check certain military targets identified by artificial intelligence before striking. One of the publication's sources described the way the document was edited as “death by a thousand small cuts“.
The document is still a draft, not a working international treaty. In November, countries are due to meet again in Geneva to discuss whether to give the talks a formal mandate and whether the process could lead to a legally binding agreement.
Warnings about “life-or-death“ decisions
According to The Washington Post, the discussions took place against the backdrop of the Pentagon's drive to quickly incorporate artificial intelligence into its military strategy. The publication writes that the US military has already used Anthropic's artificial intelligence to detect and destroy nearly 1,000 targets in the first 24 hours of the conflict with Iran.
“Machines will be able to make life-and-death decisions without human control“, says Verity Coyle, deputy director for arms issues at Human Rights Watch.According to her, weakening the rules could increase the harm to civilians and accelerate the transition to “riskier, automated warfare“.
How the current project came about
In January 2023, the US Department of Defense updated the DOD Directive 3000.09. It requires commanders and operators to exercise an appropriate level of human judgment when using autonomous and semi-autonomous weapons, as well as to comply with the laws of war and applicable international treaties.
In August 2024, UN Secretary-General António Guterres identified algorithmic targeting as a moral boundary that should not be crossed and called for a binding international treaty to be concluded by 2026. In December of that year, the UN General Assembly adopted its first resolution on lethal autonomous weapons. Only Russia, Belarus, and North Korea voted against it.
Sources: Ukrainska Pravda, Washington Post