Scholarlink

<p>This study examines whether large language models exhibit systematic country-contingent differential treatment in financial fraud detection. Analyzing 30,000 synthetic transactions with identical statistical properties across three country attributions (United States, Great Britain, and China), we find LLMs assign significantly higher fraud probabilities to Chinese-attributed transactions (36.2%) compared to Western countries (≈30–31%), resulting in accuracy disparities of 67% versus 74%. The gap remains stable across five independent experimental replications and persists when using Chinese language prompts, ruling out linguistic effects. Bias mitigation strategies, such as requiring explanations or explicit country neutrality instructions, reduce but fail to eliminate these disparities. Testing across alternative models (Llama, DeepSeek) confirms the persistence of these effects across different LLMs. These results provide evidence that an LLM’s learned associations can override objective data, highlighting the need for systematic auditing to ensure equitable outcomes in cross-border AI applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Impartial Intelligence? Evidence of Country-Label Sensitivity in AI Financial Analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月1日

查看全文
  • Fabio Motoki,
  • Jedson Pinto

摘要

This study examines whether large language models exhibit systematic country-contingent differential treatment in financial fraud detection. Analyzing 30,000 synthetic transactions with identical statistical properties across three country attributions (United States, Great Britain, and China), we find LLMs assign significantly higher fraud probabilities to Chinese-attributed transactions (36.2%) compared to Western countries (≈30–31%), resulting in accuracy disparities of 67% versus 74%. The gap remains stable across five independent experimental replications and persists when using Chinese language prompts, ruling out linguistic effects. Bias mitigation strategies, such as requiring explanations or explicit country neutrality instructions, reduce but fail to eliminate these disparities. Testing across alternative models (Llama, DeepSeek) confirms the persistence of these effects across different LLMs. These results provide evidence that an LLM’s learned associations can override objective data, highlighting the need for systematic auditing to ensure equitable outcomes in cross-border AI applications.