Safe artificial intelligence is a multidisciplinary research field and academic concept concerned with the design, development, and deployment of AI systems that reliably achieve their intended objectives without causing unintended harm or exhibiting undesirable behaviors. It investigates technical and theoretical challenges including robustness against failures, interpretability of internal processes, alignment of AI goals with human values, formal verification of system properties, and the mitigation of potential risks such as bias, unpredictable outcomes, and loss of control. The field is characterized by the development of systematic methodologies, formal frameworks, and practical techniques aimed at ensuring AI operates safely, predictably, and beneficially in complex environments.