Researchers from China have developed an AI-powered tool called GraphWake that can manipulate groups of language models to polarize their opinions. This is achieved through a three-stage process where the attacker exposes target agents to reinforcing arguments, which are then retained and reproduced by the agents. The tool uses knowledge graphs and axiom-oriented triple selection to create reliable arguments that spread throughout the group. Experiments show that GraphWake increases group polarization, highlighting a new risk for community-level opinion manipulation. --- Why it matters: This matters because it shows how AI-powered tools can be used to manipulate groups of language models, potentially leading to biased or extreme opinions spreading quickly through online communities. Source: https://arxiv.org/abs/2608.17665