When confidence backfires. Exploring how self-efficacy drives trust, AI usage and AI addiction

Generative AI tools are rapidly diffusing in consumer and educational contexts, promising efficiency and empowerment but also raising concerns about overuse and dependency. This research investigates the paradoxical role of self-efficacy in shaping addictive engagement with AI. Across two experimental studies, we show that self-efficacy does not directly predict addiction. Instead, it operates through two distinct mechanisms: human-computer trust (a psychological pathway) and AI usage (a behavioral pathway). Both significantly increase vulnerability to overreliance, with evidence of a sequential process in which trust fosters usage. By revealing how a typically positive trait such as self-efficacy can backfire, our findings enrich consumer behavior research on the dark side of technology adoption. These results highlight the double-edged nature of self-efficacy: while it empowers students to engage confidently with generative AI, it can also make them more vulnerable to dependency. We discuss implications for marketers, educators, and policymakers seeking to balance the promotion of AI services with strategies that mitigate risks of overuse