Researchers have conducted a survey on the use of deterministic methods in financial AI systems. They found that deep neural networks and Generative AI introduce mechanical nondeterminism due to hardware and architecture limitations. The study analyzed three types of models: tabular models, graph networks, and Large Language Model (LLM)-based workflows. Experiments were conducted on public financial datasets to quantify explanation rank instability, prediction flip rates, and output divergence. A layered evaluation framework was proposed to link modality-specific metrics to audit readiness. --- Why it matters: This study matters because it highlights the need for deterministic methods in financial AI systems to ensure reproducibility and transparency. Engineers working on these systems will need to consider the limitations of their models and develop strategies to mitigate nondeterminism. Source: https://arxiv.org/abs/2605.23955