Problem

The use of large language models in digital forensics lacks explainability, reducing their legal usability. Recent reasoning language models provide a chain of thought mechanism that could enhance explainability.

Method

The study used OpenAI's gpt-oss-20b model to conduct four digital forensics tasks and evaluated both the reasoning process and final answers using quantitative metrics and qualitative analysis.

Results

The chain of thought mechanism was found to be useful at medium reasoning levels but often limited. Higher reasoning levels did not enhance response quality.

Implications

The chain of thought can assist in understanding model outputs but is insufficient for explainable AI. It should be used as an aid to understanding and verifying final answers.