US and Russia strip human oversight from a UN killer-AI draft
The United States and Russia stripped human-oversight language from a landmark U.N. draft on lethal autonomous weapons during Geneva talks that wrapped earlier this month, The Washington Post's Pranshu Verma reports Sep. 26.
A lethal autonomous weapon is a system that can pick and hit targets with little or no person in the loop. Human oversight here means a person still signs off before a strike. The Post says the two nations weakened a range of safeguards in the effort, and that the negotiations marked the furthest progress yet toward a potential treaty.
Early this month, hundreds of diplomats in Switzerland worked toward the first agreement governing killer artificial intelligence arms. On the last day of negotiations, the talks turned aggressive.
CryptoBriefing's Diego Almada Lopez reports the Convention on Certain Conventional Weapons Group of Governmental Experts finished its final negotiating session on Sep. 5 after meeting in Geneva from Aug. 31. In a closed final session, roughly 20 lawyers from the U.S. and Russia produced a dramatically altered text.
Requirements for human review of AI-identified targets before a strike is authorized were removed. Provisions on predictability and reliability of AI systems in weapons contexts went with them, along with broader ethical considerations. The final document narrowed international-law scope to international humanitarian law alone, leaving out human rights law and other frameworks.
129 states participated, including 128 High Contracting Parties to the CCW. At least 76 states, including 70 CCW parties, had expressed support for the core elements that were removed or weakened as a foundation for future legally binding negotiations.
Stop Killer Robots condemned the revisions. The International Committee of the Red Cross, an observer, also criticized them.
The CCW operates by consensus, which in practice gives any single country veto-like power over outcomes. The next formal opportunity is the CCW Seventh Review Conference in November 2026.
On The Wire's Des Okoro, writing Sep. 6, called the agreed document non-binding after a three-year push. The United States and Russia held out against legally binding rules. Stop Killer Robots executive director Nicole van Rooijen said work of the past three years was "substantially diluted in the last hours," with the definition of an autonomous weapon and civilian-protection measures weakened.
Summit branding around frontier AI did not put a person back in the kill chain the draft had just dropped.

Pingback: The Sunday AI brief: Washington and Moscow thinned a killer-AI draft, OpenAI kept tool-use training paused, and Google’s Andhra campus cleared 2.51GW