Record-Locked Reproducibility Agent for Brisbane In-House Counsel: Make AI-Assisted Expert Evidence Reproducible on Demand
The board wants the deal closed by Friday. The expert report annexed to the affidavit was prepared with AI assistance — a model summarising 4,000 pages of operational data, a prompt your external solicitors drafted, a context window pulled from your own document management system. Six months later, opposing counsel asks the expert to reproduce the output. The model has been deprecated. The prompt isn’t logged. The context window has been overwritten. You can’t reconstruct the artefact, and the expert’s evidence is now exposed. The Record-Locked Reproducibility Agent exists to make sure that conversation never happens.
Why it matters now
The Australian Solicitors’ Conduct Rules (ASCR) frame solicitors’ duties as officers of the court — including the duty of candour, the duty not to mislead, and the obligation to ensure that material put before the court is capable of being substantiated. When AI is used to produce or shape evidence — whether that’s an expert report, a damages model, a document review summary, or a contractual interpretation memo — the practitioner remains accountable for what was produced and how. In-house counsel sitting between the business, external lawyers and expert witnesses are increasingly the party that has to answer the reproducibility question, because they are the only party with a continuous view across the whole evidence chain. Without a locked record of model, prompt and context, an AI-assisted output cannot be reproduced under cross-examination, cannot be re-verified after a model update, and cannot be defended against an attribution challenge.
The 5-minute view
- Expert evidence AI attribution failure is the situation where an AI-assisted output annexed to or relied on in court cannot be reproduced because the model version, prompt, or context inputs were not captured at the time of generation
- Solicitors’ duties under the ASCR — particularly duties of candour and the obligation not to mislead the court — extend to AI-assisted material put before the court
- Once a frontier model is deprecated or silently updated, an output generated against the older version is no longer reproducible unless the run was recorded at the time
- Prompts drafted in chat interfaces, context pulled from constantly-changing document stores, and outputs copied into Word are the three most common points where the audit chain breaks
- The Record-Locked Reproducibility Agent captures model identifier and version, full prompt text, every context document hash, system parameters, and the raw output — sealed into a single record at the moment of generation
- Each record gets a content-addressed identifier so the same hash can be cited in an affidavit and re-verified years later
- The agent does not store proprietary content in third-party services — records sit in your environment, under your retention policy
What Exegesis is building
The Record-Locked Reproducibility Agent is a T3 deliverable in the Exegesis Legal stack, specified in the Agentic Solutions catalogue. It wraps any AI-assisted workflow your in-house team or external advisors run — expert report generation, document review, contract analysis — and produces a sealed record of every generation event. The record contains: model name and version string, full prompt (system + user), hashes of every context document supplied, output token-for-token, timestamp, and the identity of the operator who invoked the run. Records are content-addressed so the hash itself is the citation: an affidavit can reference the record ID, and a third party can later verify that the artefact tendered matches the record. The agent is designed to integrate with the same authority-verification approach used in RuleCheck by Exegesis (rulecheck.onrender.com), so AI-assisted citations in any recorded output can be independently checked against AustLII before lodgement.
The deliverable
- A sealed reproducibility record for every AI-assisted generation event in scope
- Captured fields: model + version, full prompt, context document hashes, parameters, raw output, timestamp, operator identity
- Content-addressed record IDs suitable for citation in affidavits and expert reports
- Re-verification interface: paste a record ID and the agent returns the original artefact and verification status
- Retention configurable to match your matter-file retention policy
- Records held in your environment — not transmitted to external model providers beyond what the underlying generation already required
Why this matters in Brisbane
Queensland adopted the Australian Solicitors’ Conduct Rules in June 2012, and they bind every solicitor practising in the state — including those holding restricted in-house practising certificates. Brisbane in-house teams operating across resources, infrastructure, financial services and listed-entity governance are routinely the legal owners of expert evidence in commercial disputes, regulatory investigations and Takeovers Panel matters. When that evidence is AI-assisted, the duty of candour under the ASCR does not transfer to the external firm or the expert — it sits with whoever is responsible for the material reaching the court. The Record-Locked Reproducibility Agent is built so that responsibility is discharged with a record that can be produced on demand, years after the underlying model has changed.
Sources
- Law Council of Australia — Australian Solicitors’ Conduct Rules: https://lawcouncil.au/policy-agenda/regulation-of-the-profession-and-ethics/australian-solicitors-conduct-rules
Exegesis capability references:
Join the waitlist
The Record-Locked Reproducibility Agent is in the build queue for the Exegesis Legal stack. We’re scoping pricing around volume of generation events and retention duration rather than seats, because that’s how the risk actually scales. Join the waitlist and tell us how your team uses AI-assisted evidence today — what we hear will shape the rollout.