["Journal of Applied Philosophy, EarlyView. ", "\nABSTRACT\nData companies such as Facebook use machine learning algorithms to analyze users' personal data, revealing fine‐grained patterns about when individuals are most vulnerable. These practices, widely described as manipulative, have generated a growing literature on how data companies wrong users through their influence. In this article, I argue that existing accounts overlook a significant wrong. I show that these data companies engaged in a distinctive form of deception at the point of data collection: users were led to believe that recommendation systems existed to personalize their experience, when in fact the primary purpose was to discover their vulnerabilities. I consider several candidate accounts of what makes this deception wrong, including appeals to inauthenticity, informed consent, deception about commercial intent, and domination, and argue that none adequately captures the core wrong. I then argue that this deception constitutes a form of treachery: a deception whose purpose is to acquire power over the victim by co‐opting their agency. Drawing on cases from other contexts, I show that treachery is a pro tanto wrong that is independent of whether the acquired power is ever exercised. The concept of treachery thus identifies a prior wrong that existing accounts had not previously identified.\n"]