Across law, education, policy analysis and public moral debate, LLM outputs are increasingly used for tasks that require interpretations backed by textual evidence and explicit normative standards. A recurring failure mode, termed interpretive misplacement, occurs when model‑generated readings are treated as settled meanings without an explicit interpretive frame—no sources, scope constraints, or normative commitments—while also discarding defensible alternatives and provenance that would let readers locate supporting passages. This creates not only factual error risk but also a loss of accountability: readers and institutions cannot reliably assess what an output commits them to or on what basis.
Drawing on philosophical hermeneutics, the paper identifies this risk and derives design principles for human‑AI co‑interpretation. It first synthesises recent scholarship on hermeneutics and AI, organising it into recurring argument lines and design‑relevant gaps. LLM outputs are positioned as candidate readings; true hermeneutic understanding is reserved for accountable human interpreters situated within disciplinary historical‑linguistic traditions. Human‑AI interaction is characterised as an AI‑mediated interpretive loop, distinguishing hermeneutic understanding from token‑prediction‑based text generation. Existing LLM techniques are reorganised into design patterns that enforce framing, scope delimitation, normative mapping and provenance tracking for responsible use in interpretive settings.
The discussion then turns to implications for legal practice, educational assessment and feedback, scholarly knowledge production, and public moral argumentation. Digital hermeneutics is presented as a literacy: the ability to read AI‑mediated texts by examining frames, provenance and readings, and to contest the outputs.
Review