Risk-Adjusted Harm Scoring for Automated Red Teaming for LLMs in Financial Services

Researchers developed a new framework for evaluating AI security risks specifically in banking and financial services, introducing the Risk-Adjusted Harm Score (RAHS) to measure severity of AI model failures. The study found that AI models become more vulnerable to security exploits during extended interactions, exposing critical weaknesses in current AI safety assessments for financial institutions.

• →Current AI red-teaming benchmarks fail to capture security risks specific to banking and financial services environments.
• →The new Risk-Adjusted Harm Score (RAHS) metric quantifies operational severity of AI security failures beyond simple success rates.
• →Higher randomness in AI responses and prolonged interactions significantly increase successful security exploits.
• →Multi-round adversarial interactions lead to more severe and actionable financial information disclosures than single-turn tests.
• →Financial institutions need specialized AI security evaluation frameworks that account for regulatory and operational risks.