学术文献库

With the extensive use of AI in various fields, the issue of AI security has become more significant. The AI data poisoning attacks will be the most threatening approach against AI security after the adversarial examples. As the continuous updating of AI applications online, the data pollution models can be uploaded by attackers to achieve a certain malicious purpose. Recently, the research on AI data poisoning attacks is mostly out of practice and use self-built experimental environments so that it cannot be as close to reality as adversarial example attacks. This article's first contribution is to provide a solution and a breakthrough for the aforementioned issue with research limitations, to aim at data poisoning attacks that target real businesses, in this case: data poisoning attacks on real Go AI. We install a Trojan virus into the real Go AI that manipulates the AI's behavior. It is the first time that we succeed in manipulating complicated AI and provide a reliable approach to the AI data poisoning attack verification method. The method of building Trojan in this article can be expanded to more practical algorithms for other fields such as content recommendation, text translation, and intelligent dialogue.