Expert Report Provenance Agent for Sydney Litigation Lawyers: Know Which Paragraphs of Your Expert Report Came From a Model
Your expert is a forensic accountant. They’ve delivered a 60-page report ahead of an Administrative Review Tribunal hearing. The expert mentioned, almost in passing, that they used a generative AI tool to “tidy up” some sections. You don’t know which sections. You don’t know whether the AI rewrote conclusions, paraphrased authorities, or just smoothed grammar. The expert signs a declaration that the opinions are their own. You sign off as the instructing solicitor. If the other side asks how the report was prepared, you need a defensible answer paragraph-by-paragraph — not a hand-wave.
The problem
Expert witnesses are increasingly using LLM tools to draft, edit and structure their reports. The Administrative Review Tribunal’s expert evidence guidance and the equivalent Federal Court rules require expert reports to set out the expert’s own reasoning and opinions, with the factual basis disclosed. When a section of a report has been generated or substantially rewritten by a model, three things become unstable: (1) whether the wording still reflects the expert’s reasoning rather than a model’s plausible substitute, (2) whether cited material in that section is accurate, and (3) whether the expert can speak to the section under cross-examination. The instructing solicitor carries the consequences. Under the Australian Solicitors’ Conduct Rules — particularly Rule 19 on candour to the court and tribunal — the solicitor cannot knowingly put forward material that misrepresents its origin. The Federal Court’s General Practice Note on AI (GPN-AI) reinforces that practitioners are responsible for verifying AI-assisted content in documents filed.
The practical gap: there is no easy way to look at a finished expert report and see which paragraphs were human-authored, which were model-drafted, and which were model-edited.
What the Expert Report Provenance Agent does
The Expert Report Provenance Agent tracks the provenance of every section of an expert report — what was AI-generated, what was human-edited, and what was human-authored from the start. It produces a provenance map keyed to the structure of the report, so that for each paragraph (or numbered section) the instructing solicitor and the expert can see:
- whether the text originated with the expert
- whether a model produced an initial draft that the expert then revised
- whether a model rewrote or restructured human-authored text
- whether the section contains any citations or quoted authorities that require independent verification
The output is a provenance report you can keep with the matter file and produce if the report’s preparation is challenged.
How it works
- Ingest the working drafts. The expert (or their assistant) provides the sequence of drafts produced during preparation — initial human notes, any prompt-and-response logs from AI tools used, intermediate revisions, and the final report.
- Section the report. The agent breaks the final report into addressable units (numbered paragraphs, headings, tables, opinions, citation blocks).
- Compare against the draft history. Each unit is compared against the draft history to classify its provenance: human-authored, AI-drafted-then-edited, AI-rewritten, or unchanged-from-AI.
- Flag citation-bearing sections. Any section containing case references, statutory citations or quoted authority is flagged for independent verification — these are the highest-risk units if model-generated.
- Produce the provenance report. A structured report is returned, suitable for archiving alongside the expert’s signed declaration and the brief.
The agent does not assess the merits of the expert’s opinion. It records how the document was assembled.
Why this matters in Sydney
Sydney litigation lawyers brief expert witnesses across a wide range of ART-reviewable decisions — migration, NDIS, veterans’ affairs, taxation — as well as Federal Court and NSW Supreme Court proceedings. The ART’s practice directions and guidance for professionals set out expectations for how expert evidence is to be prepared and presented. The Federal Court’s GPN-AI sets out parallel expectations for AI use in proceedings. NSW Supreme Court expert witness codes of conduct require experts to certify the opinions are their own.
What’s changed is not the rule, it’s the workflow. Experts who would never have asked a junior to ghost-write their conclusions will routinely paste a draft section into a chatbot to “make it read better.” That edit is not visible in the final report. When opposing counsel asks the expert in cross-examination how the report was prepared, the absence of a provenance record is the problem you want to avoid. A retained provenance map — produced at the time the report was finalised, not reconstructed under pressure — is the defensible answer.
Sources
- Administrative Review Tribunal — Practice Directions and Other Guidance for Professionals and Practitioners: https://www.art.gov.au/help-and-resources/professionals-and-practitioners/practice-directions-and-other-guidance
- Federal Court of Australia — Use of Generative Artificial Intelligence Practice Note (GPN-AI): https://www.fedcourt.gov.au/law-and-practice/practice-documents/practice-notes/gpn-ai
- Law Council of Australia — Australian Solicitors’ Conduct Rules: https://lawcouncil.au/policy-agenda/regulation-of-the-profession-and-ethics/australian-solicitors-conduct-rules
- AustLII (Australasian Legal Information Institute): https://www.austlii.edu.au/
Exegesis capability references:
Join the waitlist
We’re scoping the right delivery model — per-report, per-matter, or firm licence — based on what Sydney litigators tell us about their expert briefing volumes. Tell us how you brief experts and we’ll shape access around that.