Adversarial Social Epistemology for Assemblies of Humans and Large Language Models

2026-07-11 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, AI Ethics & Societal Impact · Depth: Expert, quick

Summary

Mihnea C. Moldoveanu and Joel A.C. Baum introduce Adversarial Social Epistemology (ASE), a framework designed to analyze communicative environments where public assertions are built upon complex chains of testimony, inference, institutional certification, and implicit trust. This 50-page paper, published on arXiv on 8 Jul 2026, addresses situations where agents intentionally distort, omit, fabricate, or strategically under-specify information for personal, reputational, rhetorical, or material gain. The authors argue that existing concepts like epistemic bubbles or misinformation diffusion do not fully explain how agents exploit the inherent trustworthiness of scaffolded communications. They provide a new analytical language, detail mechanisms that subvert trust, and propose auditing machinery to redress trust breaches by enhancing the auditability of inferential chains, drawing on epistemic networks and inferentialist semantics.

Key takeaway

For AI Scientists and Research Scientists developing or deploying Large Language Models in interactive systems, you should integrate the principles of Adversarial Social Epistemology. This framework offers a robust lens to anticipate and mitigate intentional trust subversion within human-AI communication, moving beyond simple misinformation detection. Consider implementing auditing mechanisms that enhance the transparency and traceability of inferential chains to build more resilient and trustworthy systems.

Key insights

Adversarial Social Epistemology analyzes how trust is subverted in complex human-AI communication landscapes.

Principles

• Trust in public assertions relies on scaffolded chains.
• Agents exploit trust for private gains.
• Traditional models inadequately capture trust subversion.

Method

The paper outlines mechanisms that subvert trust and proposes machinery for auditing and redressing trust breaches by leveraging epistemic networks and inferentialist semantics.

In practice

• Audit inferential chains in human-LLM systems.
• Identify strategic information distortion tactics.

Topics

• Adversarial Social Epistemology
• Large Language Models
• Trustworthiness
• Epistemic Networks
• Information Distortion
• AI Ethics

Best for: AI Scientist, Research Scientist, AI Ethicist

Related on AIssential

• The Impossibility of Eliciting Latent Knowledge — The paper proves it's impossible to guarantee AI honesty about latent knowledge using only feedback on agent behavior.
• Where Does Social Reasoning Come From? Capability Provenance in Language Models — Training-data attribution reveals distinct corpus regions supporting social versus STEM reasoning in LLMs.
• Rift: A Conflict Signature for Deception in Language Models — Deception in language models leaves a detectable internal "conflict signature" distinct from honest error.
• Trust: The Hidden Scaling Mechanism of AI Systems — Trust is an operational mechanism that reduces AI system uncertainty, evaluation, and compute, enabling scalable execution.
• The Internet of Probability — LLMs transform the internet into a probability machine, eroding truth and provenance through statistical generation and contaminated training data.
• Epistemic Constitutionalism Or: how to avoid coherence bias — Large language models require explicit "epistemic constitutions" to regulate belief formation and address biases like source attribution.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.