U.S., Russia lead removal of AI weapons safety provisions in UN talks on regulation of lethal autonomous weapons

(Seoul = Yonhap News) Reporter Kang Hoon-sang = The Washington Post (WP) reported on the 26th (local time) that the United States and Russia played a leading role in deleting key provisions for the safe operation of artificial intelligence (AI) weapons in the UN's negotiations on the regulation of lethal autonomous weapons (LAW).

According to a report by WP citing anonymous sources, U.S. and Russian diplomats intensively revised the text for about 15 hours at the end of UN negotiations held in Geneva, Switzerland, earlier this month, deleting the phrase that AI weapon systems must be operated in a 'predictable and reliable' manner.

The WP reported that provisions mandating ethical considerations when using AI weapons and requiring human review before attacking military targets selected by AI were also deleted.

Three anonymous sources with knowledge of the negotiations said that the United States and Russia each brought in about 10 legal experts during a closed-door meeting where cameras were turned off and civil society observers left.

It is said that they mobilized almost twice as many manpower as other delegations and quickly revised the text, and smaller delegations had difficulty responding.

One source expressed, “The agreement slowly died due to the accumulation of numerous minor revisions.”

The U.S. State Department, Russian Foreign Ministry, and United Nations did not respond to WP's inquiries.

The UN's lethal autonomous weapons negotiations are considered the most advanced discussion in the international community to regulate how AI-based weapons are used on the battlefield. Although it is currently not legally binding, it could develop into a binding international treaty if member states reach agreement.

The countries are scheduled to meet again in Geneva in November to discuss whether to transform the negotiations into a formal delegation process or a legally binding treaty.

Verity Coyle, deputy director of crises, conflict and weapons at the human rights group Human Rights Watch (HRW), said she worries that weakened rules could allow machines to make life-or-death decisions without human control.

He pointed out, “Civilian casualties will increase, accountability will weaken, and we may quickly move towards a more dangerous automated war.”

“The United States and Russia do not support legally binding rules on autonomous weapons,” said Nicole van Loeen, secretary-general of the Campaign Against Killer Robots, an international non-governmental organization coalition, criticizing the United States and Russia for “exercising excessive influence on the outcome of negotiations.”

Military uses of AI are already becoming a reality. WP reported that the U.S. military used Antropic's chatbot to identify about 1,000 attack targets in the first 24 hours of the Iran war. Israel is also known to rely heavily on AI in its military operations in the Gaza Strip.

The U.S. government is also revising its regulations on the use of autonomous weapons. U.S. President Donald Trump issued a national security memorandum in June requiring the Department of Defense to update its autonomous weapons guidelines within 90 days, but the new guidelines have not yet been made public.

The existing guidelines created by the Joe Biden administration in 2023 include provisions that require human judgment to be applied to the inspection of autonomous and semi-autonomous weapon systems and that operators of AI weapons must comply with the laws of war and related treaties. WP pointed out that this is similar to the provisions deleted from the UN negotiation text.

“When the United States seeks to preserve its military and technological edge, it pushes for voluntary guidelines rather than legally binding rules,” Coyle told WP. “We have used the same approach in cyber, space and other arms control debates, and autonomous weapons are no exception.”

A lethal autonomous weapon refers to a weapon system in which AI can identify and track a target by analyzing sensor and video information, and determine whether to attack or perform part or all of the weapon operation. A key issue for the international community is whether such a system can prevent people from making life-or-death decisions without meaningful control.