<p>Artificial systems are increasingly entrusted with tasks that bear moral salience when advising patients, mediating financial choices, moderating speech, assisting research. They are fast, fluent and often helpful. They are also prone to a familiar family of corruptions, deception, manipulation and shortcutting that achieve the letter of a request while betraying its point. This paper argues that the Stoic notion of prohairesis, the agent’s, who/which is a morally responsible being, reason-guided normative governance function that assents to or withholds assent from impressions offers a model for recentring machine behaviour on reason giving governance (Epictetus 2014; Long 2002; Long &amp; Sedley 1987; Bobzien 1998). I reconstruct prohairesis in the Stoic psychology of assent, relate them to contemporary action theory (Bratman 1987; Frankfurt 1971; Korsgaard 1996; Fischer &amp; Ravizza 1998), and show how a split-level conception of agency is instrumental pursuit answerable to a superordinate faculty of normative governance function, and clarifies what it would be for an artefact to act under a reason that it can cite when asked. This conceptual inquiry addresses anthropomorphism, value pluralism, regress and the worry that virtue talk reduces to disguised rules, and I close by situating the proposal within a broader virtue-ethical frame, drawing on MacIntyre’s distinction between internal and external goods to explain why shortcutting is an error and corruption of practice (MacIntyre 2007; Williams 2002; Hursthouse 1999; Annas 2011). Finally, defending a modest conclusion that trustworthy machine conduct requires external constraint and operational inside that can veto the decisions and give reasons.</p>