"to perform a lethal function in a weapon without human authorization or control."
In other words... Just make sure you have a confirm dialog before launching the rockets!
IMO the major cloud players have no ethical or moral controls in place that are actually practical or useful in curtailing the usage of AI methods for the purposes of war or surveillance. It's inevitable at this point. The OpenAI/Microsoft partnership will inevitably lead to contracts with NSA/CIA/DoD. AWS/Microsoft have no real ethical pushback on DoD relationships. Google at least seems to have some internal political opposition and moral compass. The future will be a weird place.