Risk-Adjusted Harm Scoring for Automated Red Teaming for LLMs in Financial Services
Fabrizio DiminoBhaskarjit SarmahStefano Pasquali
Sep 2026
Artificial Intelligence
Abstract
Existing LLM safety evaluations rely on binary attack-success rates and domain-agnostic taxonomies, leaving regulated Banking, Financial Services, and Insurance (BFSI) deployments exposed to failures elicited through legally or professionally plausible framing. We introduce RAHS (Risk-Adjusted Harm Score), a risk-sensitive metric jointly capturing disclosure severity, disclaimer mitigation, and inter-judge agreement, and FinRedTeamBench, a 989-prompt benchmark spanning seven BFSI risk areas and 34 sub-categories mapped to regulatory frameworks. Evaluation uses an ensemble of three heterogeneous LLM judges, validated against human experts, and an adaptive multi-turn red-teaming pipeline. On nine open-weight models, RAHS preserves separation under near-ceiling ASR, ranking is stable under hyperparameter sweeps, and multi-turn pressure drives not only more jailbreaks but more operationally severe disclosures, exposing failure modes that single-turn, domain-agnostic evaluations cannot reveal.
GAOKAO-Bench is introduced, an intuitive benchmark that employs questions from the Chinese GAOKAO examination as test samples, including both subjective and objective questions that contribute a robust evaluation benchmark for future large language models and offers valuable insights into the advantages and limitations of such models.
Xiaotian Zhang, Chun-yan Li, Yi Zong et al.· arXiv.org· 216 citations· ⚡17
This work investigates the possibilities of using LLMs in a resume screening setting via a document retrieval framework that simulates job candidate selection and finds that the MTEs are biased, significantly favoring White-associated names in 85% of cases and female-associated names in only 11.1% of cases.
Empirically, PRISM reduces the end-to-end time for data selection and model tuning to just 30% of conventional pipelines, and achieves this efficiency while simultaneously enhancing performance, surpassing models fine-tuned on the full dataset across eight multimodal and three language understanding benchmarks.
Jinhe Bi, Yifan Wang, Danqi Yan et al.· arXiv.org· 73 citations· ⚡4
The method, ECCOLA, is presented, which aims at making the high-level AI ethics principles more practical, making it possible for developers to more easily implement them in practice.
Ville Vakkuri, Kai-Kristian Kemell, P. Abrahamsson· EUROMICRO Conference on Soft...· 64 citations· ⚡6
This paper designs Markov decision processes (MDPs) for different combinatorial problems and proposes to train conditional GFlowNets to sample from the solution space and demonstrates that GFlowNet policies can efficiently find high-quality solutions.
Dinghuai Zhang, H. Dai, Esmeralda S. Whitammer et al.· Advances in Neural Informati...· 59 citations· ⚡8
An empirical study on the current state of practice in artificial intelligence ethics is conducted by means of a multiple case study of five case companies, which indicates a gap between research and practice in the area.
Ville Vakkuri, Kai-Kristian Kemell, Joni Kultanen et al.· arXiv.org· 56 citations· ⚡6
