Using LLMs without trust: the adoption–trust paradox as institutional governance failure in academic ecosystems

Abstract

Large language models have rapidly made their way into higher education. Still, the integration of these tools remains contested. This study presents the results from a survey of 80 higher education professionals, most of them in Nordic countries, examining LLM adoption, trust, academic integrity, and institutional preparedness. The picture that came back is contradictory. In this self-selected sample, 82.5% already use LLMs in their teaching work. At the same time, they do not really trust what these tools produce. Trust in LLM outputs scored only 2.45 out of 5. Concerns about privacy scored considerably higher (3.82/5) and so did concerns about academic integrity (3.42/5). The respondents are fairly confident that students use LLMs regularly (4.19/5), but they have little confidence in telling student work from LLM-generated output (2.58/5). The lowest score across all areas (mean 1.99/5) was found for use of LLMs as a tool for grading students. When asked about challenges, 47.5% named academic integrity, indicating vulnerability within institutional assessment practices rather than individual misconduct. The data are consistent with a self-reinforcing cycle. Educators in this sample report clear benefits and assume widespread student use, while trusting the outputs little, and continued use does not appear to build that trust. We call this the adoption-trust paradox. The survey did not measure institutional governance directly, but Nordic policy studies report institutional responses ranging from comprehensive AI guidelines to little more than references to existing integrity norms. Read against that background, the same pattern repeated itself across every section of the survey. High adoption, low trust, and educators left to draw the lines themselves. The sample is modest and weighted toward the social sciences and humanities, but the consistency of the pattern makes it hard to dismiss, and it supports the case for an institutional response.