Artificial intelligence is an unavoidable asset of Industry 4.0. Artificial actors participate in real-time decision-making and problem solving in various industrial processes, including planning, production, and management. Their efficiency, as well as intelligent and autonomous behavior is highly dependent on the ability to learn from examples, which creates new vulnerabilities exploited by security threats. Today's disruptive attacks of hackers go beyond system's infrastructures targeting not only hard-coded software or hardware, but foremost data and trained decision models, in order to approach system's intelligence and compromise its work. This paper intends to reveal security threats which are new in the industrial context by observing the latest discoveries in the AI domain. Our focus is data poisoning attacks caused by adversarial training samples and subsequent corruption of machine learning process.