US and Russian diplomats gut AI safeguards from autonomous weapons treaty

The two countries most invested in military AI just teamed up to make sure nobody regulates it. In the final hours of a United Nations weapons negotiation in Geneva, diplomats from the United States and Russia systematically stripped the most consequential provisions from a draft framework meant to govern lethal autonomous weapons systems.

What happened in Geneva

The Convention on Certain Conventional Weapons Group of Governmental Experts, known as the CCW GGE, wrapped up its final negotiating session on September 5, 2026, after meeting in Geneva from August 31. On the last day, a closed session involving roughly 20 lawyers from the US and Russia produced a dramatically altered text.

Gone were requirements for human review of AI-identified targets before a strike is authorized. Gone, too, were provisions addressing the predictability and reliability of AI systems in weapons contexts, along with broader ethical considerations about deploying machine intelligence in warfare.

The draft had previously referenced a wide spectrum of international law. The final document narrowed that scope to international humanitarian law alone, leaving out human rights law and other frameworks that would impose additional constraints on how autonomous systems can legally be used.

129 states participated in the negotiations, including 128 High Contracting Parties to the CCW. That makes what happened on the final day particularly striking: a pair of closed-door lawyers representing two countries reshaped a document that scores of nations had spent three years building.

Who is pushing back

The backlash was immediate. Stop Killer Robots, a coalition of civil society organizations that has long campaigned for legally binding controls on autonomous weapons, condemned the revisions as a major setback. The International Committee of the Red Cross, which has observer status at CCW negotiations, also criticized the outcome.

The CCW GGE process on lethal autonomous weapons has been grinding along since 2014. Three years of the current negotiating format produced a document that, in its strongest form, represented a genuine attempt to establish baseline norms. Losing the human-review requirement in particular matters enormously: that provision was the clearest mechanism for keeping a person in the decision loop before an AI system authorizes lethal force.

At least 76 states, including 70 CCW parties, have expressed support for the core elements that were removed or weakened as a foundation for future legally binding negotiations. The CCW operates by consensus, which in practice gives any single country veto power over outcomes. The US and Russia did not need to outvote anyone. They simply needed to insist, and the text changed.

Why this moment matters

Non-binding frameworks matter because they establish norms, and norms shape behavior even without enforcement mechanisms. When the world’s two largest military powers jointly remove human-oversight language from a document about AI-controlled weapons, they are signaling something about how they intend to operate, regardless of what any treaty says.

The next formal opportunity to advance this conversation is the CCW Seventh Review Conference, scheduled for November 2026. Review conferences carry more weight than expert group sessions and could theoretically authorize a new negotiating mandate with a stronger legal character.

For the 76-plus states that supported stronger provisions, the calculus now involves a difficult choice: continue working within a consensus framework that allows two powerful states to dilute outcomes, or pursue a separate treaty process outside the CCW entirely, as has happened with landmines and cluster munitions.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.