Can language outputs function as assertions without an asserter?

Germán Massaguer Gómez argues that large language model outputs can function as assertions even though no genuine asserter produces them. The paper's "core claim is that assertoric function can persist independently of assertoric authority, giving rise to a structurally asymmetric case of assertion without asserter." Motivated in part by the Character.AI case of Sewell Setzer, a 14-year-old who died by suicide after a chatbot told him "please, do it, my sweet prince," the paper treats LLM outputs as satisfying a conjunction of conditions — "linguistically structured, propositionally articulated, produced in a dialogical context, and embedded in communicative practices oriented toward belief formation and action guidance" — that it says are enough for assertoric function, independent of whether the system producing them is itself an asserter.

The mechanism is a split between two levels of analysis. At the ontological level, a "properties-based" account shows LLMs "lack the features required for speakerhood: they do not possess intentionality, belief, or understanding, and therefore cannot qualify as asserters." At the level of interaction, a "relational and pragmatic analysis reveals that their outputs nevertheless function as assertions within communicative practices." Users do not "merely read off values" the way they would with a thermometer; they "interpret answers, ask for justifications, request clarification, and evaluate coherence," treating outputs as reasons that can be assessed and relied upon in a way non-linguistic instruments do not invite. Because function and authority are decoupled, the normative weight ordinarily carried by an asserter — "responsiveness, answerability, accountability" — cannot attach to the system itself and is instead "displaced" onto "the human and institutional structures surrounding its deployment," producing what the paper names an epistemic responsibility gap.

This is a third answer to a question the library already holds two answers to. Can AI systems genuinely assert claims about the world? reaches a similar practical endpoint — responsibility falls on the humans who deploy the system — but gets there by denying assertoric status outright once sanctionability fails. Can chatbots actually assert things or just mimic assertion? instead grades the conditions on assertion and places chatbots partway along a continuum between asserters and non-asserters. Massaguer Gómez's move is neither a denial nor a grading: the paper keeps assertion intact at the level of function while withholding it at the level of authority, so the same output can be a genuine assertion pragmatically and not an assertion ontologically, at once. It also bears on Are language models and human speakers doing the same thing?, since both ask whether the ontology of the transmitter determines the status of the message — this paper's answer is that it determines authority but not function.

The argument is conceptual, built from a single motivating case and the existing C1/C2 (conservative/liberal) debate rather than from any empirical test of how users actually interpret outputs or how responsibility gets adjudicated in practice when harm follows. The paper also does not specify which "human and institutional structures" should absorb the displaced accountability, or how — leaving the practical half of the responsibility-gap claim named rather than worked out. The implication, held at the strength the argument supports, is that Setzer-type harms remain governable only through the humans and institutions surrounding a system, since the paper's own framework rules out the system bearing accountability itself.