Paper proposes adversarial social epistemology framework for human–LLM assemblies

Researchers propose an adversarial social epistemology (ASE) framework to analyze how communicative agents—including humans and large language models—exploit the commitments and entitlements that normally make scaffolded public assertions trustworthy. The authors argue that existing concepts such as epistemic bubbles, echo chambers, and misinformation diffusion do not fully capture the dynamics of trust subversion in densely interactive communicative landscapes.

The paper introduces a conceptual language to describe how agents may distort, color, omit, fabricate, or strategically under-specify information for private, reputational, rhetorical, or material gains. It focuses on how such behaviors exploit the scaffolding of public assertions—chains of testimony, inference, institutional certification, and tacit trust—rather than merely amplifying or diffusing misinformation.

Mechanisms that subvert trust in scaffolded public communications are outlined, alongside machinery for auditing and redressing trust breaches. The proposed approach draws on epistemic networks and an inferentialist semantics for interpreting assertions, aiming to make inferential chains auditable and to detect when their auditability has been compromised.

The work is presented as a conceptual and analytical contribution rather than an empirical evaluation, positioning ASE as a lens for studying information integrity in human–AI assemblages.