AI Forensic Investigation Reference Matrix
AFIRM
An openly licensed knowledge base for post-incident forensic investigation of AI systems. Rows are AFIRM's seven AI evidence classes, ordered by asserted volatility (EC01 most volatile → EC07 most stable). Columns are the five forensic phases.
A technique renders in a cell only if it directly consumes or produces that evidence class in that phase; cross-cutting techniques render as a phase band below the grid instead of being duplicated into every row.
| Evidence class | P01 Identification | P02 Collection | P03 Examination | P04 Analysis | P05 Reporting |
|---|---|---|---|---|---|
| EC01 Live model & session state |
|
|
|
| — |
| EC02 Runtime telemetry & interaction logs |
|
|
|
| — |
| EC03 Retrieval & memory stores |
|
| — |
| — |
| EC04 Deployment configuration |
|
| — |
| — |
| EC05 Model artifacts at rest |
|
|
| — | |
| EC06 Training & pipeline artifacts |
|
| — |
| — |
| EC07 Documentation & static records |
| — | — | — | — |
Phase bands
Cross-cutting techniques
These techniques range over an entire phase rather than a single evidence class, and are rendered here instead of duplicated into every cell above.
- P01 — Identification
- T0101 Incident Characterization, T0102 Access Level Determination, T0103 System Composition Mapping
- P02 — Collection
- T0208 Chain of Custody for AI Artifacts
- P04 — Analysis
- T0406 Contributing-Factor Assessment, T0407 Incident Timeline & Narrative Reconstruction
- P05 — Reporting
- T0501 Structured Claim Formulation, T0502 Evidentiary Standard Assessment, T0503 Regulatory Reporting Alignment