Moral Framing of Military Artificial Intelligence: The Role of Trust and Attitudes in Defensive and Offensive Use

Abstract

Despite growing research on public perceptions of Artificial Intelligence (AI), limited attention has been paid to how moral framing shapes evaluations of military AI. Drawing on research on trust in AI and the halo effect, this study examines whether offensive versus defensive framing influences users’ perceptions across multiple dimensions, including cognitive attitudes, affective attitudes, and perceived intelligence. The study employed an online questionnaire with a between-subjects design. Participants were randomly assigned to one of two conditions, in which they were presented with a scenario describing the development of an AI algorithm for either defensive or offensive military purposes. Results showed that, compared to defensive framing, offensive framing significantly reduced both cognitive and affective attitudes toward the AI system, as well as its perceived intelligence. Trust in AI emerged as a strong and independent predictor across outcomes. Structural equation modeling indicated that the effect of framing on perceived intelligence was largely mediated by affective and cognitive attitudes, consistent with a halo-like process in which moral evaluations shape competence judgments. No moderating effects of identification with all humanity were observed; personality traits were not significant predictors of attitudinal evaluations. However, identification with all humanity showed a significant negative association with affective attitudes. These findings suggest that the perceived intelligence of military AI does not reflect an independent assessment of system capability, but emerges from cognitive, affective and moral evaluations triggered by its intended use. More broadly, they highlight how relatively subtle differences in framing can systematically influence public perceptions, raising concerns about the potential use of defensive narratives as persuasive devices in the communication of military AI. This study contributes to HCI by showing that AI evaluations are not only context-dependent but also morally constructed, with important implications for the design, communication, and governance of AI in high-stakes domains.