A new method for detecting coalitions in multi-agent AI systems has been introduced, focusing on analyzing internal neural representations rather than just observable behaviors. This approach enhances the understanding of emergent organizational structures, which is crucial for AI safety and alignment.