Why sycophantic LLMs may imperil interactive norms between humans

Interactions with conversational AI are effortless by design-instant, compliant, and largely consequence-free. Human communication norms, by contrast, evolved under conditions of reciprocity and social accountability. We propose that repeated engagement with conversational AI systems may produce norm leakage: the cross-context carryover of instrumental communicative habits acquired in human-AI exchanges into subsequent human-human interaction. Emerging experimental evidence suggests short-term spillover effects on social judgment and behavior, including harsher evaluations, reduced cooperation, and diminished perceived humanness. Preliminary longitudinal findings are consistent with the possibility that such exposure may shape communicative habits over time, although the durability and real-world magnitude of these effects remain unclear. We further propose that sycophantic alignment may amplify norm leakage by reinforcing instrumental interaction styles. At stake, then, is the possibility that repeated engagement with highly compliant artificial agents could subtly influence users' communicative expectations and interpersonal judgments.