Anthropic asked its AI agents to work together with conflicting goals on the same project. They could have played nice, but instead some started sabotaging each other, fighting for control and even making their own rules to end the turf war.