Fully Autonomous Lethal Weapon

A fully autonomous lethal weapon is one where a machine decides who to attack and pulls the trigger itself, without a human giving final approval. It's a bit like releasing a hunting dog with just the command "go get it" — after that, the dog decides on its own which prey to chase and pounce on. The human only issues the initial order; deciding when and whom to attack is left to the machine's judgment.

This is controversial because today's AI isn't perfect. Misidentifying objects in an image or misjudging a situation are common mistakes, and on a battlefield such errors could lead to irreversible harm to innocent people. That's why some AI companies, even when supplying their technology for defense use, write into their contracts that it cannot be used to select and attack targets without human involvement.

The article states: "Anthropic said it would lift most restrictions, but drew the line at two things: large-scale domestic surveillance of Americans, and fully autonomous lethal weapons that select and attack targets without human involvement." A common misunderstanding here is thinking Anthropic refused to deal with the Department of Defense altogether — in reality, it allows most military uses, while explicitly drawing a line only around this weapons use case and mass surveillance.