The use of algorithms to predict behaviour is becoming the gold standard in criminal justice in various countries. This article critically analyses the algorithm-driven risk assessment tools used in predictive policing and predictive justice. First, we propose to see algorithms as essentially bureaucratic instruments. They are the digital offspring of the classic bureaucratic procedure, creating classification through standardised and impersonal decision-making. Second, we argue that the application of algorithms in criminal justice expands the bureaucratic field to areas previously understood as bulwarks of professional judgement. Third, we analyse the shift in purpose of algorithmic decision-making: instead of determining a citizen's status of beneficiary or obligate, we now see algorithmic anticipation of behaviour. This shifts the logic of decision-making over investigations, probations, and sentencing from individual judgement to bureaucratic classification based on the algorithms that are designed into risk assessments tools. This article is both a bureaucratic critique of algorithm-driven risk assessment tools in criminal justice and a call to rethink bureaucracy and bureaucratisation beyond the boundaries of public administration.