New Framework Analyzes Trust Subversion in AI-Human Communication

Researchers have proposed a new framework called Adversarial Social Epistemology (ASE) to understand how trust can be undermined in communication landscapes where humans and large language models (LLMs) interact extensively. This framework goes beyond traditional concepts like echo chambers or misinformation, focusing instead on how agents strategically manipulate the commitments and entitlements that typically make shared assertions reliable. The paper details specific mechanisms through which trust in public communications, built on chains of testimony, inference, and institutional certification, can be subverted. It provides a specialized language for this analysis and outlines methods for auditing and rectifying trust breaches. The proposed machinery draws on epistemic networks and an inferentialist semantics to interpret assertions, aiming to restore accountability in inferential chains.