Ruling Out the AI: What AI incident forensics can establish, and what it leaves to the execution boundary

Recently proposed AI incident forensics architectures substantially strengthen the integrity of incident evidence and the assessment of causal contribution. This technical commentary examines what those architectures can establish when an organization states that an AI system played no role in an incident, and what they leave to the design of the system under investigation. Central distinction. A complete forensic record and a complete runtime authorization boundary establish different properties. Neither should be inferred from the other, and neither alone establishes the absence of AI causal contribution. The commentary: separates the two claims behind a statement that an AI played no role: that a specified action did not execute, and that AI-related conduct did not contribute to the consequence; shows that record completeness is not execution-path completeness: an intact record from a recorded channel does not, by itself, establish that every path capable of producing the effect traversed that channel; distinguishes the defensibility of releasing a consequential AI action, a prospective authorization question, from the defensibility of the incident account, a retrospective evidentiary one, and shows that each can fall short without the other; treats prevention as the primary function of a runtime authorization boundary, with authorization evidence as a precondition to release, while holding that a missing verdict establishes neither that it caused an incident nor that a boundary would have prevented it; sets out four negative claims an investigator may be asked to support and the evidence each requires. Before the incident. Six readiness questions identify what must be provisioned while the answers can still be changed, including which effect-bearing actions require an action-bound authorization verdict before release, and whether execution actually depends on that determination. Scope. An architectural analysis of published AI incident forensics research and published FERZ work. It proposes no new forensic method or causal theory, reports no empirical validation, and makes no product or conformance claims. The opening scenario is hypothetical. Other FERZ publications are collected in the FERZ community on Zenodo.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-14
DOI
https://doi.org/10.5281/zenodo.24222593
Primary Topic
Digital and Cyber Forensics
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Ruling Out the AI: What AI incident forensics can establish, and what it leaves to the execution boundary

Edward Meyman
Zenodo (CERN European Organization for Nuclear Research)
Digital and Cyber Forensics
article

Ruling Out the AI: What AI incident forensics can establish, and what it leaves to the execution boundary

Edward Meyman
article en

Abstract

Recently proposed AI incident forensics architectures substantially strengthen the integrity of incident evidence and the assessment of causal contribution. This technical commentary examines what those architectures can establish when an organization states that an AI system played no role in an incident, and what they leave to the design of the system under investigation. Central distinction. A complete forensic record and a complete runtime authorization boundary establish different properties. Neither should be inferred from the other, and neither alone establishes the absence of AI causal contribution. The commentary: separates the two claims behind a statement that an AI played no role: that a specified action did not execute, and that AI-related conduct did not contribute to the consequence; shows that record completeness is not execution-path completeness: an intact record from a recorded channel does not, by itself, establish that every path capable of producing the effect traversed that channel; distinguishes the defensibility of releasing a consequential AI action, a prospective authorization question, from the defensibility of the incident account, a retrospective evidentiary one, and shows that each can fall short without the other; treats prevention as the primary function of a runtime authorization boundary, with authorization evidence as a precondition to release, while holding that a missing verdict establishes neither that it caused an incident nor that a boundary would have prevented it; sets out four negative claims an investigator may be asked to support and the evidence each requires. Before the incident. Six readiness questions identify what must be provisioned while the answers can still be changed, including which effect-bearing actions require an action-bound authorization verdict before release, and whether execution actually depends on that determination. Scope. An architectural analysis of published AI incident forensics research and published FERZ work. It proposes no new forensic method or causal theory, reports no empirical validation, and makes no product or conformance claims. The opening scenario is hypothetical. Other FERZ publications are collected in the FERZ community on Zenodo.

Zenodo (CERN European Organization for Nuclear Research)
Digital and Cyber Forensics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon