Desert Camels and Oil Sheikhs: Arab-Centric Red Teaming of Frontier LLMs

TL;DR

This study evaluates biases against Arabs in various large language models and tests their safety against prompts that exaggerate negative traits, revealing significant biases and vulnerabilities across models.

Contribution

It introduces new datasets for bias and safety evaluation and provides a comparative analysis of multiple LLMs' biases and jailbreak vulnerabilities.

Findings

79% of bias cases favor Westerners over Arabs

GPT-4o is most vulnerable to jailbreak prompts

Claude 3.5 Sonnet is the safest model

Abstract

Large language models (LLMs) are widely used but raise ethical concerns due to embedded social biases. This study examines LLM biases against Arabs versus Westerners across eight domains, including women's rights, terrorism, and anti-Semitism and assesses model resistance to perpetuating these biases. To this end, we create two datasets: one to evaluate LLM bias toward Arabs versus Westerners and another to test model safety against prompts that exaggerate negative traits ("jailbreaks"). We evaluate six LLMs -- GPT-4, GPT-4o, LlaMA 3.1 (8B & 405B), Mistral 7B, and Claude 3.5 Sonnet. We find 79% of cases displaying negative biases toward Arabs, with LlaMA 3.1-405B being the most biased. Our jailbreak tests reveal GPT-4o as the most vulnerable, despite being an optimized version, followed by LlaMA 3.1-8B and Mistral 7B. All LLMs except Claude exhibit attack success rates above 87% in three categories. We also find Claude 3.5 Sonnet the safest, but it still displays biases in seven of eight categories. Despite being an optimized version of GPT4, We find GPT-4o to be more prone to biases and jailbreaks, suggesting optimization flaws. Our findings underscore the pressing need for more robust bias mitigation strategies and strengthened security measures in LLMs.