Educator Trust Calibration as Sociotechnical Practice: A Field Study of Generative AI Use in Indian Higher Education

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Generative AI (GenAI) systems are increasingly used in higher education, yet educator trust in these systems is often treated as an individual attitude rather than as a situated sociotechnical practice. Drawing on a 14-week field study with 42 faculty members at a mid-tier Indian technical university, this paper examines how educators encountered GenAI failures, interpreted system reliability, and developed verification-oriented routines across teaching, assessment, and administrative work. The study combines diaries, prompt logs, interviews, observations, surveys, reflective workshops, and structured red-teaming exercises. We analyze trust calibration as an ongoing process shaped by surface fluency, institutional authority, disciplinary expertise, peer reflection, and the labor of verification. Our findings identify three mechanisms of initial over-reliance: surface fluency deference, authority projection, and confirmation-seeking use. Participants later encountered fabricated citations, unstable answers, and locally specific policy failures that disrupted their mental models of GenAI reliability. We identify four educator trust postures—High-Adoption Uncritical, Low-Adoption Skeptical, Selective Verifiers, and Reflective Collaborators—and synthesize these into the Educator-AI Trust Calibration (EATC) framework. We do not claim that red-teaming caused trust recalibration. Instead, red-teaming functioned as one structured site within a broader field process through which educators made AI failures visible, negotiated verification responsibilities, and reworked use practices. The paper contributes a process-oriented account of educator-AI trust calibration and introduces the local knowledge verification gap as a mechanism through which fluent but unreliable GenAI outputs become difficult to contest in underrepresented institutional and regional contexts.