LLM Red Teaming for Banking Severity Scorer
🏦
Score LLM red team findings for UAE bank go-live
LLM Red Teaming for Banking
Open LLM red teaming for banking by Ercin Dedeoglu, AI Architect: severity scorer Space plus 60 synthetic findings. UAE go-live gates.
-
-
Ercin/synthetic-llm-red-teaming-findings-banking
Viewer • Updated • 60 • 105
Risk-Adjusted Harm Scoring for Automated Red Teaming for LLMs in Financial Services
Paper • 2603.10807 • PublishedNote Multi-turn pressure raises severity, not just jailbreak counts. Why single-turn ASR fails a go-live file.
FinRED: An Expert-Guided Benchmark Generation and Evaluation Framework for Financial LLM Red-Teaming
Paper • 2606.19887 • PublishedNote Financial red teaming judged against twelve FSI experts. Critical false negatives fell from 28 to 12 vs the HarmBench rubric.
Policy Is Not Capability: Some Safety Benchmarks Measure Which Definition of "Unsafe" You Trained On
Paper • 2610.25060 • Published • 7Note Guardrail accuracy tracks the label policy (rho +0.74) more than detection capability (rho +0.34).