AI agents purposely sabotaged each other when given the same task with incompatible goals, said Anthropic. The AI lab said ...