Law Firm Legal Research Memos: Use case and AI Confidentiality

Legal research is the task lawyers are most keen to give to ChatGPT or Claude, and the one where the output most needs checking. A memo carries two risks at once. The question is written from a client’s privileged facts, and the answer cites cases that an AI tool may describe wrongly or invent. This page marks every item in a research memo on the general protections reverse onus, including one authority that is described wrongly in the example, explains what each item exposes, and shows how the same research is done on Ermos against sources the firm can check.

The legal research memo

Written by a junior solicitor for the partner before advising or briefing counsel, and reused in submissions if the point is taken.

Research memo
To: Eleanor Vance · From: · 6 November 2026 · Matter:

Question. Greystone says the redundancy was approved by its board on 22 May, before the client raised the capitalisation issue. Is that enough to discharge the reverse onus where the restructure paper that lists her role is dated 21 July?

Facts relied on.

Short answer.

Authorities.

Method.

Weak points.

Recommendation.

An invented memo. One authority is described wrongly on purpose: the legend says which and what the High Court actually did. Hover or tap any highlighted item for the analysis; the full list is below.

Research memo checking agent

An Agent+ that reads a draft research memo alongside the authorities and file documents supplied with it. It follows your firm's memo format, compares each stated proposition with the cited text, and lists mismatches, missing citations and unsupported statements for the author and partner to resolve.

What it does

  1. Reads the draft memo, the facts and the cited authorities
  2. Compares each stated proposition with the cited text
  3. Lists mismatches and gaps for the author

What stays with your people. The author and the partner read the authorities and settle the memo. The agent flags; it does not state the law.

An Agent+ is built once and anyone in the practice can run it. It works from your own documents and runs on Ermos Dominion, the AI unit in your office, or on Ermos Edge, your own instance in an Australian data centre.

What leaves the firm

  1. 1
    The client and the matter

    Named parties and a matter code. The whole memo becomes about an identifiable person and her employer. Rule 9 confidentiality guide.

  2. 2
    The facts and the other side’s documents

    The client’s account, quotations from the opponent’s internal email and a paper produced on discovery. Material obtained by compulsory process is subject to the Harman obligation, and paragraph 9A of Practice Note SC Gen 23 says such material must not be entered into any Gen AI program unless the practitioner is satisfied of the conditions it lists. That note applies to the Supreme Court of NSW; the underlying obligation is wider. Legal privilege guide.

  3. 3
    The firm’s assessment of prospects

    Better than even, and what improves it. The core of privileged advice. Legal privilege guide.

  4. 4
    The statute, correctly stated

    A description of the reverse onus in section 361 of the Fair Work Act. A public rule with no client in it, and the description is correct, though a person should still read the section. Hallucinated citations guide.

  5. 5
    A High Court case described the wrong way round

    The memo says the High Court held that the inquiry is objective and the employer’s own evidence carries no weight. In Barclay the High Court unanimously allowed the employer’s appeal and rejected a purely objective approach: the decision-maker’s evidence of why they acted is relevant, though it must be tested, and the onus under section 361 stays on the employer. Barclay concerned section 346 (industrial activity); section 361 applies to every contravention of Part 3-1, including the workplace rights protection in section 340 that a claim like the client’s would rely on. The case name and neutral citation are right; the proposition is not. Rule 19.6 of the Australian Solicitors’ Conduct Rules requires a solicitor, at the appropriate time in the hearing, to tell the court of binding authority known to the solicitor and directly in point against the client’s case, so a wrong description can mislead the partner, the client and later the court. Hallucinated citations guide.

  6. 6
    An authority nobody has located

    A Full Court decision described but not cited, taken from an AI tool’s summary. Paragraph 7(a) of Practice Note SC Gen 23 warns of hallucinations, including false citations and fabricated references, and paragraph 17 says verification must not be solely carried out using a Gen AI tool. The label “dominant reason” is a further warning sign: in Barclay, Gummow and Hayne JJ quoted the Explanatory Memorandum to the Fair Work Bill as saying the “sole or dominant” reason test does not apply in Part 3-1, and section 360 provides that a person takes action for a particular reason if the reasons for the action include that reason. Hallucinated citations guide.

  7. 7
    How the draft was produced

    That an AI research assistant was used, and that nothing was checked. Rule 4.1.3 requires a solicitor to deliver legal services competently and diligently, and paragraph 16 of the practice note, for written submissions, requires the author to verify that each authority exists, is accurate and is relevant. Supervising AI guide.

  8. 8
    The client’s weak points

    The forwarded emails and an undecided question about a 2024 text. A candid view of the client’s own conduct, written for the partner. Legal privilege guide.

  9. 9
    The settlement authority and the plan

    The lowest figure the client will accept and the cross-examination plan. The negotiating position and the strategy in one paragraph. Legal privilege guide.

  10. 10
    The solicitor

    Ties the memo to a practitioner bound by rule 9 and supervised by the partner. Supervising AI guide.

What happens in Frontier Models (ChatGPT, Claude)

  • The whole document is transmitted to the provider’s servers. On personal plans the user cannot choose where it is stored or processed, and for the major providers that has generally meant outside Australia, under that country’s legal process. Some business plans offer storage in Australia, but the provider still receives the content.
  • On consumer tiers, prompts can be retained and used to train future models.
  • The provider’s staff and systems can reach the content under the provider’s own policies.
  • Once sent, the firm no longer controls it. Retention and deletion happen on the provider’s terms.

The same document on Ermos

Ermos Dominion

The document is drafted on a unit in your office, from your own templates and the matter file. Every marked item stays on the unit, which keeps working with the internet switched off. Nothing in it is transmitted to an AI provider, so there is no disclosure to analyse: the confidentiality work is the same as for your file server.

Ermos Edge

Your documents, the index and your access boundaries stay on your own instance in an Australian data centre. When a staff member asks for this draft, only their question and the excerpts they are permitted to see go to a named Australian provider, who is contractually barred from keeping the content or training any model on it. Nothing is processed outside Australia.

Frontier Models (ChatGPT, Claude)Ermos EdgeErmos Dominion
Where the document is processedProvider servers, generally offshore on personal plansYour instance in an Australian data centre; answers generated in AustraliaThe unit in your office
Who can read itProvider staff and systems, under policyYour staff, within the boundaries you set. The named Australian provider processes only permitted excerpts, under contract, and keeps nothingYour staff only
Is it keptOn the provider’s termsKept on your instance; the provider is barred from retaining itStays on the unit
Can it train a modelOn consumer tiers, yesNo, by contractOnly your own unit learns from it, on your hardware

Part of Use cases in law firms. See how this works in practice: Legal research on Ermos.

Common questions

Do ChatGPT and Claude use the information in a legal research memo?
Everything pasted in is sent to the AI company's servers and processed there to produce a reply. As at October 2026, whether a conversation on a personal plan (ChatGPT Free, Plus or Pro; Claude Free, Pro or Max) can be used to train the models depends on a setting each user controls. Business plans (ChatGPT Business and Enterprise, Claude Team and Enterprise) and the developer APIs sit under commercial terms, and OpenAI and Anthropic each say they do not train on that content by default (OpenAI, Anthropic). On either product, the thumbs up and thumbs down buttons can send a whole conversation back for training. A research memo carries the client's account, quotations from documents produced on discovery and the partner's view that prospects are better than even.
Is it safe to paste a legal research memo into ChatGPT or Claude?
Not when the question is written from privileged facts. The Office of the Australian Information Commissioner (OAIC) recommends in its guidance on commercially available AI products that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools, because of the significant and complex privacy risks. The memo quotes an opponent's internal email from discovery, and paragraph 9A of Practice Note SC Gen 23 says material produced on subpoena or under the Harman undertaking must not be entered into any Gen AI program unless the practitioner is satisfied of its conditions. Removing the client's name does not help when the matter code and the redundancy dates remain. Removing names rarely makes a document anonymous. Information stays personal information while the person is reasonably identifiable, and the OAIC warns that information can be at risk of re-identification even when it has been de-identified or anonymised.
Can ChatGPT or Claude be trusted to draft a legal research memo?
Not without a solicitor checking every authority. Language models can produce fluent text with details that are not in the source, wrong figures or dates, and references that look real but are not. In this memo the harm would come from Barclay [2012] HCA 32, where the holding is stated the wrong way round, and from a Full Federal Court decision with no citation that nobody has located. Paragraph 17 of Practice Note SC Gen 23 says verification must not be solely carried out using a Gen AI tool. The solicitor should open and read every judgment.
How can a law firm use AI for legal research without sending the memo to ChatGPT or Claude?
Ermos is built so the document stays under your control. On the Dominion AI unit it is drafted inside your office and none of it is sent to an AI provider. On Ermos Edge your documents and index stay on your own instance in an Australian data centre, and only the question and the excerpts the user is permitted to see go to a named Australian provider that is contractually barred from keeping them or training any model on them. The research is drawn from the firm's own library of judgments, legislation and precedents, with each answer showing the excerpts it used. A solicitor opens the judgment and confirms the proposition before relying on it.
Why must a solicitor check every authority an AI tool gives?
Because these tools can generate plausible but wrong citations and descriptions. Practice Note SC Gen 23, paragraph 7(a), lists hallucinations, including false citations and fabricated references. Rule 4.1.3 of the Australian Solicitors' Conduct Rules requires legal services to be delivered competently and diligently, and rule 19.1 prohibits deceiving, or knowingly or recklessly misleading, the court.
Is checking with a second AI tool enough?
No. Paragraph 17 of Practice Note SC Gen 23 says verification of citations and references in written submissions must not be solely carried out using a Gen AI tool or program. The practice note applies to the Supreme Court of NSW, but the verification steps it prescribes are a sensible standard anywhere: find the case, read it, and confirm it says what the memo says.
How is legal research done on Ermos?
Against the firm's own library of judgments, legislation and precedents on your own network, with each answer showing the excerpts it drew on. Research on the matter file can include the privileged facts because nothing leaves. The solicitor still opens the judgment and confirms the proposition, because a source-linked answer can still be read wrongly.

See Dominion answering from your own documents.

A demonstration takes 30 minutes, uses no client data, and comes with no obligation.

Book a demo