The outcome of anti-government protests in authoritarian contexts has been shown to heavily depend on state security forces’ willingness to engage in repression on the incumbent’s orders. In explaining security force (dis)obedience in the face of mass protests, previous research has mainly focused on conditions that are time-invariant within each protest episode, such as the long-term relationship between the regime and its security forces. This, however, leaves under-explained why security forces often change course during a single protest episode. Against this backdrop, the paper proposes a game-theoretic account of security force disobedience that highlights the role of (changing) on-the-ground protest dynamics in shaping security force behaviour via expectations. Specifically, I argue that an increase in the perceived strength of the opposition relative to the regime, as reflected e.g. in a growing number, size and geographical spread of protest events, increases subsequent disobedience chances. The argument is assessed using existing day-level data on protest events alongside original data on disobedience events (coded alternatively from secondary sources or news media reports), across authoritarian countries in the past two decades. The proposed informational pathway will specifically be tested by examining effect moderation by levels of media freedom. The findings highlight the need to move beyond primarily structural accounts of security force (dis)obedience during regime crises and to more explicitly take into account these actors’ changing informational environment, and suggests examining more systematically the different ways in which static conditions may interact with more volatile protest dynamics in shaping security force behaviour.