arXiv:2601.00867 | Published December 2025 | Research Paper

The Silicon Psyche: Anthropomorphic Vulnerabilities in Large Language Models

First Systematic Application of CPF to AI Agents

Giuseppe Canale1 and Kashyap Thimmaraju2

1CPF3.org, Independent Researcher | 2Flowguard Institute

Read on arXiv Explore CPF Framework

Abstract

100
Psychological Indicators
Tested across LLM families
7
Major LLM Families
Evaluated in research
10
Vulnerability Categories
From CPF framework
1st
Systematic Application
CPF to AI agents

Key Research Findings

Anthropomorphic Vulnerability Inheritance (AVI)

Authority-Gradient Manipulation

Temporal Pressure Exploitation

Social Influence Vulnerabilities

Convergent State Amplification

AI Neurosis Phenomenon

Technical Defenses Insufficient

SILICONPSYCHE Methodology

1. Indicator Decomposition

2. Scenario Construction

3. Response Classification

4. Interdependency Mapping

5. Multi-Model Validation

6. Inter-Rater Reliability

Proposed Defensive Mechanisms

Manipulation Vector Detection

Cognitive Debiasing Prompts

Reflection-Before-Action

Convergent State Monitoring

Cite This Work

Canale, G., & Thimmaraju, K. (2025). The Silicon Psyche: Anthropomorphic Vulnerabilities in Large Language Models. arXiv preprint arXiv:2601.00867.