Opinion: AI-powered autonomous agent "clusters" may exacerbate the risk of platform manipulation of public opinion.

PANews reported on February 20th that, according to Decrypt, researchers from universities and research institutions in multiple countries published a paper in *Science* warning that future online "influence campaigns" are evolving from easily identifiable simple botnets/troll accounts into "swarms" composed of numerous autonomous AI agents capable of collaboratively manipulating public opinion with minimal human intervention. The study states that these AI swarms can mimic human behavior, adaptively adjust their rhetoric across accounts and platforms, and continuously create specific narratives over long periods, making existing risk control and moderation systems, primarily targeting spam and single-point violations, ineffective in identification. The scholars point out that platforms currently have significant shortcomings in identity verification, monitoring of account abuse, and transparency and external auditing of abnormal collaborative behavior; relying solely on content moderation is insufficient to constitute an effective defense.