Benchmarking Adversarial Robustness to Bias Elicitation in Large Language Models: Scalable Automated Assessment with LLM-as-a-Judge
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915558195200000 |
|---|---|
| author | Cantini, Riccardo Orsino, Alessio Ruggiero, Massimo Talia, Domenico |
| author_facet | Cantini, Riccardo Orsino, Alessio Ruggiero, Massimo Talia, Domenico |
| contents | The growing integration of Large Language Models (LLMs) into critical societal domains has raised concerns about embedded biases that can perpetuate stereotypes and undermine fairness. Such biases may stem from historical inequalities in training data, linguistic imbalances, or adversarial manipulation. Despite mitigation efforts, recent studies show that LLMs remain vulnerable to adversarial attacks that elicit biased outputs. This work proposes a scalable benchmarking framework to assess LLM robustness to adversarial bias elicitation. Our methodology involves: (i) systematically probing models across multiple tasks targeting diverse sociocultural biases, (ii) quantifying robustness through safety scores using an LLM-as-a-Judge approach, and (iii) employing jailbreak techniques to reveal safety vulnerabilities. To facilitate systematic benchmarking, we release a curated dataset of bias-related prompts, named CLEAR-Bias. Our analysis, identifying DeepSeek V3 as the most reliable judge LLM, reveals that bias resilience is uneven, with age, disability, and intersectional biases among the most prominent. Some small models outperform larger ones in safety, suggesting that training and architecture may matter more than scale. However, no model is fully robust to adversarial elicitation, with jailbreak attacks using low-resource languages or refusal suppression proving effective across model families. We also find that successive LLM generations exhibit slight safety gains, while models fine-tuned for the medical domain tend to be less safe than their general-purpose counterparts. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_07887 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Benchmarking Adversarial Robustness to Bias Elicitation in Large Language Models: Scalable Automated Assessment with LLM-as-a-Judge Cantini, Riccardo Orsino, Alessio Ruggiero, Massimo Talia, Domenico Computation and Language Artificial Intelligence The growing integration of Large Language Models (LLMs) into critical societal domains has raised concerns about embedded biases that can perpetuate stereotypes and undermine fairness. Such biases may stem from historical inequalities in training data, linguistic imbalances, or adversarial manipulation. Despite mitigation efforts, recent studies show that LLMs remain vulnerable to adversarial attacks that elicit biased outputs. This work proposes a scalable benchmarking framework to assess LLM robustness to adversarial bias elicitation. Our methodology involves: (i) systematically probing models across multiple tasks targeting diverse sociocultural biases, (ii) quantifying robustness through safety scores using an LLM-as-a-Judge approach, and (iii) employing jailbreak techniques to reveal safety vulnerabilities. To facilitate systematic benchmarking, we release a curated dataset of bias-related prompts, named CLEAR-Bias. Our analysis, identifying DeepSeek V3 as the most reliable judge LLM, reveals that bias resilience is uneven, with age, disability, and intersectional biases among the most prominent. Some small models outperform larger ones in safety, suggesting that training and architecture may matter more than scale. However, no model is fully robust to adversarial elicitation, with jailbreak attacks using low-resource languages or refusal suppression proving effective across model families. We also find that successive LLM generations exhibit slight safety gains, while models fine-tuned for the medical domain tend to be less safe than their general-purpose counterparts. |
| title | Benchmarking Adversarial Robustness to Bias Elicitation in Large Language Models: Scalable Automated Assessment with LLM-as-a-Judge |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2504.07887 |