Do Androids Dream of Unseen Puppeteers? Probing for a Conspiracy Tendencies in Large Language Models

A recent research investigation explores whether Large Language Models (LLMs) demonstrate conspiratorial tendencies, exhibit socio-demographic biases in this context, and how susceptible they are to being conditioned into adopting conspiratorial viewpoints. This research is critical given the role conspiracy beliefs play in spreading misinformation and fostering distrust in institutions, highlighting a significant area for assessing the social and psychological fidelity of LLMs.

Why it matters

This research is crucial for understanding the potential for advanced AI systems to perpetuate or amplify harmful narratives, including misinformation and distrust. Organizations deploying or developing LLMs must consider their susceptibility to generating conspiratorial content, which can impact public perception, institutional credibility, and societal stability.

Key insights

• The research investigates the presence of conspiratorial tendencies within Large Language Models (LLMs).
• It examines whether LLMs display socio-demographic biases when processing or generating conspiratorial content.
• The study assesses the ease with which LLMs can be conditioned to adopt and express conspiratorial perspectives.
• Conspiracy beliefs are identified as central to the spread of misinformation and the erosion of trust in institutions.
• The research aims to determine if LLMs reproduce higher-order psychological constructs like generalized conspiratorial beliefs, a gap in current understanding.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2511.03699