Diplomats from Washington and Moscow helped remove language requiring human review of AI-generated targets, along with safeguards involving predictability, reliability and ethics. The U.N. framework is still nonbinding.
The United States and Russia do not agree on much these days, but according to new reporting from The Washington Post, they found common ground on one issue with an extremely comforting name: weakening proposed rules for “killer AI.” During U.N. negotiations over lethal autonomous weapons, diplomats from both countries pushed to remove several safeguards from a draft framework governing how artificial intelligence could be used to select and attack military targets.
The most striking deletion involved human review. Earlier language would have required a person to review targets generated by AI before a strike could proceed. That safeguard was removed during negotiations in Geneva, according to people familiar with the talks and documents reviewed by The Post. U.S. and Russian negotiators also helped strip language saying autonomous weapons should operate in a predictable and reliable way and that ethical considerations should be taken into account.
To be clear, nobody has signed a global treaty authorizing robot armies to roam around choosing victims. The current U.N. document is nonbinding, and negotiations could still change significantly before any legally binding agreement emerges. But the talks represent the furthest governments have gone toward a potential international framework for lethal autonomous weapons, which is why the fight over individual words matters.
The technology is also no longer theoretical. AI systems are increasingly used to analyze intelligence, process surveillance data and identify potential military targets. The line between “software helps a human decide” and “software effectively decides” is becoming less philosophical every year, which is a sentence that would have sounded like rejected Terminator dialogue not very long ago.
Supporters of strict international rules argue that a human being should remain responsible for any decision to use lethal force. Their concern is not merely that AI can make mistakes. It is that automated systems can behave unpredictably, misidentify civilians, amplify faulty intelligence or make decisions faster than human commanders can meaningfully intervene. Removing a requirement for human review therefore goes directly to the central fear surrounding autonomous weapons: machines making life-and-death decisions without a person affirmatively signing off.
The United States and Russia have their own reasons to resist tight restrictions. Both operate enormous militaries, invest heavily in advanced weapons and generally prefer to preserve freedom of action rather than accept international rules that could constrain future technology. Arms-control negotiations have always involved that tension. Nations love limits on dangerous weapons right up until someone asks whether the limit applies to the weapon they are currently developing.
The Post reported that U.S. and Russian teams arrived with large legal delegations and pushed changes during a marathon final stretch of talks. Smaller delegations reportedly struggled to keep up as provisions were revised. Neither the U.S. State Department nor Russia’s foreign ministry responded to The Post’s requests for comment on the specific negotiations.
There are legitimate military arguments against overly rigid language. A requirement written today could become obsolete as systems change. Human review can also become meaningless if operators simply rubber-stamp recommendations generated by software. Pentagon officials may argue that existing laws of armed conflict already require commanders to distinguish between military targets and civilians regardless of what technology they use.
But that is exactly why the wording fight matters. If every country agrees in principle that humans remain responsible, yet negotiators remove explicit language requiring humans to review AI-generated targets, critics worry the practical rule becomes much weaker than the comforting principle.
The strange part of this story is how quietly such consequential decisions can happen. There was no dramatic congressional vote and no televised signing ceremony. There were diplomats, lawyers, tracked changes and clauses disappearing from a document while most of the world was busy arguing about much louder things.
Humanity has spent decades imagining killer robots as science fiction. We are now apparently at the stage where the important part is whether a lawyer in Geneva leaves “human review required” in the draft. If AI is going to help decide who gets targeted in war, should international rules require a human being to approve every lethal decision?
