Law Firm AI Tool Procurement Agreements: Use case and AI Confidentiality

A business buying an AI tool signs a contract about what happens to everything it feeds the tool and everything the tool produces. The agreement deals with use of inputs and outputs, a promise that the vendor will not train on the customer’s data, confidentiality, ownership of outputs, audit rights, where the data sits and a disclaimer about accuracy. Staff paste the draft into ChatGPT or Claude to compare it with a checklist, and the draft carries the client’s negotiating notes. This page marks every item in an AI procurement agreement that rule 9, privilege and the Privacy Act bear on, and shows how the same draft is reviewed on Ermos inside the firm.

The AI tool procurement agreement

Prepared when a business has chosen an AI vendor and the contract is the last step before the tool is switched on.

AI Services Agreement
Draft 2 · Tuesday 10 November 2026

Parties. .

1. Customer Data.

2. No training. Customer Data is the Customer’s Confidential Information. The Vendor may use it only to provide the Service and must limit access to staff who need it.

3. Ownership of Outputs.

4. Data location.

5. Audit.

6. Accuracy.

7. Incidents.

Note to Wattle Hollow, delete before sending:

Executed by .

An invented agreement with the firm’s note still attached. The vendor and its product are invented, and the clauses are contract terms that say nothing about any real product. Hover or tap any highlighted item for the analysis; the full list is below.

AI vendor contract checker

An Agent+ that reads a vendor's draft AI tool agreement beside your firm's own clause library and the client's instructions. It checks the clauses on inputs, outputs, training, data location and audit rights against your positions, and prepares a short list of gaps and suggested wording from your precedents.

What it does

  1. Reads the vendor draft and the client's instructions
  2. Compares each clause with your precedent positions
  3. Lists gaps and proposes wording from your clause library

What stays with your people. The solicitor reviews every flagged point and negotiates. A partner approves what goes back to the vendor.

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 parties and what the tool is for

    The client, the vendor and what the client plans to feed the tool. Client information under rule 9, and a statement of a business process the client may not want known. Rule 9 confidentiality guide.

  2. 2
    Inputs and Outputs as Customer Data

    A definition that treats what the tool generates as the customer’s data as well as what goes in. The OAIC’s guidance on commercially available AI products, published on 21 October 2024, says privacy obligations apply to personal information input into an AI system and to output data that contains personal information. Privacy Act guide.

  3. 3
    The no training promise and confidentiality

    A contractual promise by the vendor about its own conduct, and the term that makes it enforceable. The OAIC’s guidance says organisations adopting a commercially available AI product should do due diligence, including on who will have access to personal information put into or generated by the product. A contract term is not a statement about any real product. Choosing legal AI tools.

  4. 4
    Ownership of Outputs

    An allocation of rights between the parties. A contract allocates whatever rights exist; it does not decide whether any right exists in an Output, which this document does not address.

  5. 5
    Data location and sub-processors

    APP 8.1 says that before an APP entity discloses personal information to an overseas recipient, it must take reasonable steps in the circumstances to ensure the recipient does not breach the Australian Privacy Principles (other than APP 1), and section 16C can treat the recipient’s breach as the entity’s. It applies to an APP entity, which a small business operator under section 6D is not unless an exception applies. Privacy Act guide.

  6. 6
    Audit rights

    The right to inspect the vendor’s controls and logs. The OAIC’s guidance points to APP 11 (reasonable steps to protect personal information from misuse, interference and loss, and from unauthorised access, modification or disclosure) when assessing a new AI product. Privacy Act guide.

  7. 7
    The accuracy disclaimer

    Standard wording that outputs may be wrong and must be checked. The OAIC says APP 10 requires reasonable steps to ensure personal information used and disclosed is accurate, including appropriate use of disclaimers that communicate limits on accuracy. The clause names no one.

  8. 8
    Incident notice within 24 hours

    A contract term with a commercial period. Part IIIC of the Privacy Act separately deals with eligible data breaches for entities the Act covers; the 24 hours is the client’s negotiated figure. Privacy Act guide.

  9. 9
    The firm’s note: what the data is and what the client will trade

    The kinds of personal and customer-confidential data the tool will see, a confidentiality duty to two other customers, and the client’s negotiating plan. Privileged advice and information that is confidential to people who are not the client. Legal privilege guide.

  10. 10
    The signatories and the solicitors

    Named directors and initials. Personal information, and a link between the client and the practitioners who advised it. 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: Agreement review on Ermos.

Common questions

Do ChatGPT and Claude use the information in an AI procurement agreement?
Whatever is 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. This draft carries Wattle Hollow Cold Chain's plan to feed delivery dockets and temperature logs to Valewood Assist, and a note that the dockets include drivers' names and licence numbers.
Is it safe to paste an AI procurement agreement into ChatGPT or Claude?
No, because the note to Wattle Hollow is privileged and records that Valewood has resisted clause 4 and that the client will trade down to named sub-processors. 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 note also says the temperature logs belong to two pharmaceutical customers whose contracts with the client are confidential, so the draft holds information that is confidential to people who are not the client.
Does a ChatGPT Business or Claude Team plan make it safe for an AI procurement agreement?
Not on its own. On a personal plan the user cannot choose where conversations are stored or processed. Some business plans let an organisation choose to store data at rest in Australia (OpenAI), which still leaves the content with the AI company. A business plan changes the default on training, but it does not change who receives the content. This agreement is about keeping Customer Data in Australia and away from overseas sub-processors, and its negotiating note would go to an AI company that is not a party to it. The OAIC's guidance also tells organisations to review carefully the terms and settings that will apply to their use of any AI product. Our comparisons with ChatGPT Enterprise and Claude weigh the trade-offs.
How can a law firm use AI for an AI procurement agreement without sending it 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 review is prepared from the draft agreement, the client's checklist and the firm's own agreement review template. A solicitor decides the positions on training, data location and audit, and checks the Privacy Act references.
What does APP 8 require when personal information goes to an overseas AI vendor?
APP 8.1 says that before an APP entity discloses personal information to an overseas recipient, it must take reasonable steps in the circumstances to ensure the recipient does not breach the Australian Privacy Principles, other than APP 1. Under section 16C, if the recipient then does something that would breach them, the entity is taken to have breached them. APP 8.2 lists exceptions, such as consent after the individual is expressly informed.
Does the Privacy Act apply to every business that buys an AI tool?
No. A business with annual turnover of $3 million or less is a small business under section 6D, and a small business operator is generally outside the Act. Section 6D(4) removes that status in some cases, including a business that provides a health service and holds health information, or discloses personal information about another individual for a benefit, service or advantage. Larger businesses are generally covered.
What does the OAIC say about commercially available AI products?
Its guidance, published on 21 October 2024 and updated on 17 January 2025, says privacy obligations apply to personal information input into an AI system and to output data that contains personal information. It recommends due diligence on suitability, human oversight and who has access, and as best practice that organisations do not enter personal information, particularly sensitive information, into publicly available generative AI tools.

See Dominion answering from your own documents.

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

Book a demo