Skip to content
Open access

From capability to assertability: epistemic regulation of LLM-mediated claim presentation

Sep 2026 · Ethics and Information Technology · Vol 28 · 0 citations · 35 references

TL;DR

This work develops assertability as a relational normative status of an output-in-practice, determined by the evidential and contextual conditions governing its presentation, across four regulatory loci—assertoric force, evidential grounding, evidential candor, and epistemic response selection.

Abstract

Improvements in large language model (LLM) capability can increase accuracy, reasoning, and access to relevant information while leaving unassessed whether a particular output is appropriately presented for epistemic uptake. This creates an evaluative gap: first-order capability metrics can register genuine gains while leaving claim-presentation conditions underdetermined. LLM outputs increasingly occupy assertion-like roles without thereby acquiring the standing of accountable asserters. We develop assertability as a relational normative status of an output-in-practice, determined by the evidential and contextual conditions governing its presentation. One such condition is the evidence frame: the set of sources admissible as grounds within an interaction. A true answer, or one well supported outside the governing frame, may therefore remain non-assertable when it relies on an inadmissible evidential basis. To address this capability–assertability gap, we use epistemic regulation to denote the normative-functional regulation of whether and how an LLM-mediated output is presented for epistemic uptake in light of the conditions bearing on its assertability. We organize the framework across four regulatory loci—assertoric force, evidential grounding, evidential candor, and epistemic response selection. The framework is normative-functional, while its realization may be distributed across learned behavior, retrieval and verification mechanisms, interface design, evaluation, deployment rules, and institutional practices, with responsibility and accountability remaining with identifiable human and organizational actors. Because standards governing assertion, qualification, verification, and withholding can distribute epistemic risk and verification burdens unevenly, their specification and oversight are also governance questions rather than merely technical calibration choices.

Read PDF

Similar papers

Open access Oct 2026

Extended Judgment: LLMs and the Architecture of Artificial Capacities

Large language models (LLMs) are increasingly embedded in everyday epistemic practices, yet their epistemological status remains unclear. Existing debates typically oscillate between treating LLMs as epistemic agents or reducing them to mere tools. This paper argues that both views mischaracterize the core issue. Dra...

J. Noller · 0 citations
Open access Oct 2026

Charity Without Competence: Domain Knowledge and the Limits of Interpretation in AI Ethics

Charitable interpretation in technically complex domains depends on accurately characterizing the object of interpretation. This article develops a domain-competence condition for evaluating arguments in artificial intelligence (AI) ethics. Domain competence is question-relative rather than synonymous with engineering...

Ammar Younas, Yi Zeng · 0 citations
Open access Sep 2026

Who asserts when no one asserts? Functionalism, proxy responsibility, and conversational ideals in the age of LLMs

Recent philosophical work on large language models (LLMs) has increasingly appealed to theories of assertion to make sense of our conversational interactions with these systems. Two strategies are especially prominent: functionalist accounts, which characterize assertion in terms of the public conversational roles and...

Germán Massaguer Gómez, Antonio Gaitán Torres · 0 citations
#human-computer interacti... Preprint Oct 2026

The Ethics of Prompt Engineering

Prompt engineering is commonly understood as a technical practice aimed at eliciting useful, accurate, and appropriately structured outputs from large language models (LLMs). Yet prompts do more than optimize model performance: they shape what information is elicited, which assumptions are introduced, and which perspec...

F. Germani, G. Spitale · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.