MANTA, a novel framework, enables multi-agent systems to adapt their network topology dynamically, allowing for more efficient communication and problem-solving. This approach deviates from traditional methods, which often treat communication topology as a fixed design element or optimize it offline. By introducing adaptability, MANTA enhances the overall performance of large language model-based multi-agent systems, particularly in complex problem-solving tasks. The framework facilitates task decomposition, agent specialization, and information exchange, leading to improved intermediate validation. As state-aligned threat activity increases, the implications of MANTA extend beyond the immediate target, raising the stakes from criminal to geopolitical1. This matters to practitioners because dynamic network topology adaptation can significantly impact the security and resilience of multi-agent systems, making MANTA a crucial consideration for those developing and deploying such systems.