Abstract: Algorithmic trading can learn methods of manipulating the securities market and, with its advantages in technical efficiency and the homogeneity of algorithms, can break through the “self-deterrence” of manipulation, even triggering risks such as flash crashes. Since artificial intelligence lacks legal personhood, both securities traders and algorithm service providers—who have control over the risks of algorithmic trading—should be deemed responsible subjects for the manipulation. Given the “autonomous nature” of algorithmic trading, traders may conceal their manipulative intent through inducement-based manipulation. Therefore, when pursuing liability, information-based supervisory tools should be utilized to infer manipulative intent from unreasonable trading patterns or from facts such as algorithmic homogeneity in training. Considering the “black box” problem of algorithms and the real risks of manipulation, negligence should be added as a fault type for manipulation liability. Moreover, a presumption of fault should be imposed on service providers, who must prove that they have fulfilled their duty of care to prevent manipulation. Including providing anti-manipulation warnings and monitoring, as well as demonstrating through model interpretation that the algorithm is free from defects.