Decoding-temperature stress testing for financial LLM security

This use case checks whether a finance model becomes more dangerous when you make its text generation more random. It compares different temperature settings to see when harmful answers become more likely and more severe.