This article examines the “digital welfare state” historically, presently, and into the future, with a focus on what artificial intelligence means for welfare surveillance. Drawing on scholarship about the development of bureaucracy, the welfare state, and automation, as well as specific examples from the Netherlands, I argue that problems posed by artificial intelligence in public administration are often misplaced or misattributed and that the societal challenges we can expect to encounter in welfare surveillance are more likely to be historically familiar than technologically novel. New technologies do provide some new capabilities, which explains the uptake of algorithmic tools in welfare fraud investigation and the use of chatbots in assisting with welfare applications. Algorithmic systems are also increasingly subject to “audits” and regulations that mandate accountability. However, many of the key issues in the automation of the welfare state are the same as identified in scholarship that long precedes the current hype around artificial intelligence. These issues include a persistent suspicion of welfare recipients to justify surveillance as a form of fraud identification, opaque decision-making, and punitive measures directed against marginalized groups, enacting harm and reproducing inequalities.