AI-assisted verification integrates large language models, image analysis tools, and automated data processing into established investigative workflows to accelerate pattern identification, translation, summarisation, and hypothesis generation. The critical constraint is that AI output requires independent corroboration before any claim is published: no finding rests on model output alone. AI tools are subordinate instruments in this workflow, not primary evidence sources.
This technique applies when a document set is too large for manual review, when audio or foreign-language material requires initial triage, or when structured patterns across datasets need to surface for follow-up verification. The AI flags candidates; the investigator confirms them. Each AI-generated output, individually, is a hypothesis. The pattern, corroborated against primary sources, is evidence.
AIV-001 sets out the workflow for selecting, deploying, and documenting AI tools in the verification workflow to evidentiary standard.
Eight workflow steps, six tooling sources, five false-positive checks, five chain-of-custody requirements.
|
01
Required Tools
Six platforms covering LLM analysis, transcription, document reasoning, and data cleaning.
|
02
OPSEC
Data transmission rules for sensitive material and cloud service boundaries.
|
|
03
Workflow
Eight-step sequence from task definition to documented case-folder record.
|
04
False Positives
Hallucination, translation error, entity misidentification, and confidence score misreading.
|
|
05
Chain of Custody
Five requirements for recording tool, prompt, hash, and verification steps per task.
|
06
Key Queries
Six operator references across Claude, Whisper, NotebookLM, OpenRefine, Diffchecker, and Azure.
|
Download the card.
A PDF version of AIV-001 is available below for Signal subscribers.
