Anthropic's issue was only that the AI isn't yet good enough to tell who's an American, so it avoids killing them. They were fine with the "killing non-Americans" bit.
> But today, frontier AI systems are simply not reliable enough to power fully autonomous weapons. We will not knowingly provide a product that puts America’s warfighters and civilians at risk.
That’s interesting: I recall reading that sentence but I parsed it as “America’s warfighters” AND “civilians” (i.e., civilians in general), rather than as “America’s warfighters and America’s civilians”. I think the former reading is more plausible since American civilians are not normally at physical risk from American military operations. It’s unfortunate that this sentence is worded ambiguously, though.
