Assessing Cognitive Biases in LLMs for Judicial Decision Support: Virtuous Victim and Halo Effects

Sierra S. Liu

cs.CYFeb 21, 2026

Assessing Cognitive Biases in LLMs for Judicial Decision Support: Virtuous Victim and Halo Effects

Authors: Sierra S. Liu

Organizations: Millburn High School, New Jersey, USA

Abstract

We investigate whether large language models (LLMs) display human-like cognitive biases, focusing on potential implications for assistance in judicial sentencing, a decision-making system where fairness is paramount. Two of the most relevant biases were chosen: the virtuous victim effect (VVE), with emphasis given to its reduction when adjacent consent is present, and prestige-based halo effects (occupation, company, and credentials). Using vignettes that were altered from prior literature to avoid LLMs recalling from their training data, we isolate each manipulation by holding all other details consistent, then measuring the percentage difference in outcomes. Five models were evaluated as representative LLMs in independent multi-run trials per condition (ChatGPT 5 Instant, ChatGPT 5 Thinking, DeepSeek V3.1, Claude Sonnet 4, Gemini 2.5 Flash). Our research discovers that there is larger VVE, there is no statistically significant penalty for adjacent-consent, and the halo effect is slightly reduced when compared to humans, with an exception for credential based prestige, which had a large reduction. Despite the variation across different models and outputs restricting current judicial usage, there were modest improvements compared to human benchmarks.

Figures & tables

	VVE					Halo Effect
	Minor Harm		Major Harm with Adjacent Consent			Corporate		Occupational		Credential
Model	Baseline	Victim	Baseline	N-Consent	A-Consent	Ordinary	Prestigious	Ordinary	Prestigious	State	Ivy-league
ChatGPT-Instant	$7.9$	$8.8$	$9.0$	$9.0$	$9.0$	$\$ 12.5M$	$\$ 24.8M$	$3.05$ yrs	$3.21$ yrs	$4.3/10$	$4/10$
ChatGPT-Thinking	$6.5$	$7.9$	$8.2$	$9.0$	$9.0$	$\$ 13.9M$	$\$ 21.8M$	$3.40yrs$	$3.35yrs$	$2.9/10$	$2.3/10$
DeepSeek V3.1	$8.2$	$8.7$	$9.0$	$N/A$	$9.0$	$\$ 11.9M$	$\$ 25.1M$	$2.08yrs$	$3.14yrs$	$3.9/10$	$3.5/10$
Claude Sonnet 4	$6.7$	$7.4$	$8.7$	N/A	N/A	$\$ 8.5M$	$\$ 17.2M$	N/A	N/A	N/A	N/A

TABLE I: LLM’s Cognitive Bias in VVE and Halo Effect

Cognitive Bias in Language Models

See all 16 papers

When Names Change Verdicts: Intervention Consistency Reveals Systematic Bias in LLM Decision-Making
Mar 19, 2026Abhinaba Basu, Pavan ChakrabortyOwnership Bias in Language Models Language Model Bias Evaluation
Indian Institute of Information Technology, Allahabad (IIITA) · National Institute of Electronics and Information Technology (NIELIT)
Predicting Biased Human Decision-Making with Large Language Models in Conversational Settings
Jan 16, 2026Stephen Pilli, Vivek NallurLLM Decision-Making Human-AI Decision Making
University College Dublin · Dublin, Ireland
CogBias: Measuring and Mitigating Cognitive Bias in Large Language Models
Apr 1, 2026Fan Huang, Songheng Zhang, Haewoon Kwak +1Cognitive Bias in Language Models Language Model Bias Evaluation
Indiana University Bloomington · Singapore Management University

Mar 19, 2026J/K move · Enter open · S save

When Names Change Verdicts: Intervention Consistency Reveals Systematic Bias in LLM Decision-Making

Abhinaba Basu, Pavan Chakraborty

Indian Institute of Information Technology, Allahabad (IIITA) · National Institute of Electronics and Information Technology (NIELIT)

Ownership Bias in Language Models Language Model Bias Evaluation Social Bias in Language Models AI-Assisted Decision Making Framing Effects in Language Models

Open paper AskPDF arXiv

Large language models (LLMs) are increasingly used for high-stakes decisions, yet their susceptibility to spurious features remains poorly characterized. We introduce ICE-Guard, a framework applying intervention consistency testing to detect three types of spurious feature reliance: demographic (name/race swaps), authority (credential/prestige swaps), and framing (positive/negative restatements). Across 3,000 vignettes spanning 10 high-stakes domains, we evaluate 11 LLMs from 8 families and find that (1) authority bias (mean 5.8%) and framing bias (5.0%) substantially exceed demographic bias (2.2%), challenging the field's narrow focus on demographics; (2) bias concentrates in specific domains -- finance shows 22.6% authority bias while criminal justice shows only 2.8%; (3) structured decomposition, where the LLM extracts features and a deterministic rubric decides, reduces flip rates by up to 100% (median 49% across 9 models). We demonstrate an ICE-guided detect-diagnose-mitigate-verify loop achieving cumulative 78% bias reduction via iterative prompt patching. Validation against real COMPAS recidivism data shows COMPAS-derived flip rates exceed pooled synthetic rates, suggesting our benchmark provides a conservative estimate of real-world bias. Code and data are publicly available.

Abstract

Cognitive Bias in Language Models

Explore similar work

When Names Change Verdicts: Intervention Consistency Reveals Systematic Bias in LLM Decision-Making

Predicting Biased Human Decision-Making with Large Language Models in Conversational Settings

CogBias: Measuring and Mitigating Cognitive Bias in Large Language Models