Recognition, Simulation, and Refusal: A Contamination-Aware Study of Classic Psychological Effects in LLM Agents
Summary
This paper investigates the recognition, simulation, and refusal of classic psychological effects in LLM agents using a contamination-aware methodology.
View Cached Full Text
Cached at: 09/22/26, 08:59 AM
# Recognition, Simulation, and Refusal: A Contamination-Aware Study of Classic Psychological Effects in LLM Agents Source: [https://arxiv.org/abs/2609.22090](https://arxiv.org/abs/2609.22090) Bibliographic Tools ## Bibliographic and Citation Tools Bibliographic Explorer Toggle Code, Data, Media ## Code, Data and Media Associated with this Article Demos ## Demos Related Papers ## Recommenders and Search Tools About arXivLabs ## arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website\. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy\. arXiv is committed to these values and only works with partners that adhere to them\. Have an idea for a project that will add value for arXiv's community?[**Learn more about arXivLabs**](https://info.arxiv.org/labs/index.html)\.
Similar Articles
State Contamination in Memory-Augmented LLM Agents
This paper identifies and studies 'memory laundering' in LLM agents, where toxic or adversarial context compressed into memory summaries evades standard toxicity detectors while still influencing future generations. It introduces the sub-threshold propagation gap (SPG) to measure hidden downstream influence and shows that sanitizing toxic state before summarization is more effective than post-hoc cleaning.
How Affect Propagates among LLM Agents: Emergent Emotional Contagion in Crowd Simulation
This paper presents a multi-agent crowd simulation using LLM-driven agents that perceive and appraise each other through visual, auditory, and tactile channels, leading to emergent emotional contagion without explicit hand-authored affect transfer. The study demonstrates spatial, temporal, and personality-dependent contagion dynamics across five scenarios and evaluates backend-dependent appraisal behavior.
Illusory Truth or Mere Exposure? Model-Dependent Repetition Effects in LLM-Based Social Media Simulations
This University of Zurich paper investigates whether the Illusory Truth Effect replicates across four LLMs (Gemma-3-4b-it, Qwen2.5-7B-Instruct, Llama-3.1-8B-Instruct, GPT-5-nano) in generative agent-based social media simulations, finding qualitatively distinct repetition effects — including genuine ITE, mere exposure, and no effect — cautioning against assuming uniform bias replication in LLM-based social simulations.
CAPTURE: Disentangling Preference Drift from Memory Poisoning in Personalized LLM Agents
The paper introduces CAPTURE, a method to distinguish between genuine user preference changes and malicious memory poisoning in personalized language agents using neural differential equations and causal auditing.
Thinking Less to Simulate Better: Intuitive Prompting Improves LLM Agents Simulating Individual Social Media Reactions, Including Unfamiliar Content
This study shows that intuitive prompting improves LLM agents' simulation of individual social media reactions, achieving higher fidelity than analytical methods, even for unfamiliar content.