Tag
This paper introduces a spectral diagnostic method to detect hidden coalitions in multi-agent AI systems by analyzing internal neural representations via mutual information, addressing critical AI safety and alignment challenges.