This thesis examines how the use of artificial intelligence, AI, is legitimized in relation to public authorities’ decision-making processes in Swedish public documents. Rather than assuming that AI is legitimate or illegitimate in itself, the study focuses on how public actors justify different uses of AI and automation, particularly when considerations of efficiency, legal certainty, accountability and control intersect.The empirical material consists of strategically selected Swedish public reports, guidelines and policy-related documents concerning AI, automation and decision support. The study is conducted as a qualitative document study using thematic content analysis. Scharpf’s concept of output legitimacy is used to analyse arguments about expected results and problem-solving capacity. Process-related legitimacy requirements are used in this thesis as an analytical category for demands concerning legal certainty, transparency, documentation, accountability and control. Van Leeuwen’s legitimation strategies are used to examine how these arguments are constructed through rationalization, moral evaluation and authorization.The findings show that utility is the most prominent legitimation argument, but its weight changes with the role of the technology. Efficiency, productivity and capacity are especially important when AI is discussed as administrative support. When AI moves closer to decisions affecting individuals, legal certainty, accountability, transparency and meaningful control become more important. Risk is also used in two directions: to justify caution and oversight, but also to justify continued development when inaction is presented as a risk. The overall conclusion is therefore that AI is legitimized conditionally rather than in generalf. Legitimacy rests not only on what the technology is expected to achieve, but also on whether its use can be assessed, monitored, corrected and, when necessary, limited.