American and Russian diplomats removed several safeguards from a draft United Nations text on lethal autonomous weapons during talks in Geneva earlier this month, including a provision that would have required humans to review AI-selected targets before a strike, The Washington Post reported Saturday. The report, by Pranshu Verma, was republished by The Spokesman-Review.

What Happened

According to the Post, the changes were made over roughly 15 hours on the final day of the September negotiating round. The U.S. and Russian teams struck three sets of language from the draft: a requirement that AI weapons systems operate in a predictable and reliable way, a clause calling for ethical considerations to be weighed when such weapons are used, and the requirement for human review of machine-generated targets.

The work took place in a closed-door session, with U.N. cameras switched off and civil-society observers removed from the room, the paper reported. Each of the two delegations fielded about 10 lawyers, which the Post described as nearly double the presence of other countries. One negotiator characterized the process as "death by a thousand paper cuts."

RelatedU.S. Begins Selecting Leaders for Program Enlisting Firms Against Foreign Cybercriminals

The U.S. State Department, the Russian Foreign Ministry and the United Nations did not respond to the Post's requests for comment.

Why It Matters

Human review of targeting decisions has been the central demand of arms-control advocates for as long as states have debated autonomous weapons. Verity Coyle, a deputy director at Human Rights Watch, told the Post the weakened text could leave machines making life-and-death decisions without human control. Nicole van Rooijen of the Stop Killer Robots campaign said the two governments have the leverage to block progress that most states want toward a binding agreement.

The episode also shows how far Washington's position has moved. The Post noted that a Pentagon directive issued under the Biden administration in 2023 required human judgment over autonomous weapons and compliance with the laws of war. In June, President Trump issued a national security memorandum ordering that directive rewritten within 90 days; as of late September no new version had been published. Trump has said his administration rejects any globalist effort to control AI, and the Pentagon is moving to build the technology quickly into battlefield planning.

The paper cited recent military use as context, reporting that the U.S. military used an AI chatbot to help identify nearly 1,000 targets in the first 24 hours of its conflict with Iran, and that Israel has relied heavily on AI in Gaza.

Glass-walled European institution building
Illustrative photo: Unsplash. Not a photograph of the reported event.

Context

Talks on lethal autonomous weapons have run for years in Geneva under the Convention on Certain Conventional Weapons, a forum that works by consensus, which lets any single state hold up a final text. The Post described the September round as the furthest the effort has advanced, which is why changes made in its last hours carry weight: the draft is the basis for any future instrument.

For technology companies that sell AI models and software to defense customers, the outcome shapes what obligations may eventually follow their products into military use. A treaty requiring human review of targets would set a baseline that procurement rules and export policies could reference; a text without it leaves those questions to national policy.

What to Watch

States reconvene in Geneva in November to decide whether to pursue a formal mandate for a binding treaty. Key questions are whether other delegations try to restore the deleted provisions, whether the U.S. and Russia hold together on blocking them, and when the Pentagon releases its rewritten autonomous weapons directive, which will signal how much human control the United States intends to keep over its own systems.