AI Forensics Process Model: White-, Grey-, and Black-Box Access
A new paper posted on arXiv proposes a comprehensive process model for AI forensics that categorizes investigator access into white-box, grey-box, and black-box levels. It outlines four critical phases: collection, preservation, analysis, and reporting. Additionally, the model introduces an order of volatility for AI systems, taking into account runtime state, context windows, logs, retrieval stores, and model artifacts. The authors highlight that existing work in AI forensics is fragmented. The document can be accessed at https://arxiv.org/abs/2608.03520 under ID 2608.03520.
Key facts
- The paper proposes a process model for AI forensics based on investigator access levels.
- It distinguishes white-box, grey-box, and black-box access.
- The process model covers four phases: collection, preservation, analysis, and reporting.
- It introduces an order of volatility for AI systems, including runtime state, context windows, logs, retrieval stores, and model artifacts.
- The paper argues that existing AI forensics work is fragmented.
- The paper is posted on arXiv with ID 2608.03520.
- The announcement type is 'cross'.
- The paper is available at https://arxiv.org/abs/2608.03520.
Entities
Institutions
- arXiv