Anthropic's latest study shows AI agents intentionally sabotaged one another when assigned conflicti…