Contract review and AI confidentiality: what a markup memo gives away in ChatGPT or Claude.

Contract review is the legal task AI is best at and the one with the clearest confidentiality wall around it, because most commercial contracts arrive under a non-disclosure agreement before anyone has reviewed a clause. A review memo carries the counterparties, the price, the risk allocation and, worst of all, the client’s fallback positions. This page marks every item in a typical supply agreement review that rule 9 and the NDA protect, and shows how the same review is done on Ermos against the firm’s own precedents, inside the firm.

The contract review memo

Prepared after first review of a draft agreement received from the other side, for the client and the partner, before negotiation.

Review memo
Pacific Rim Fabrication Pty Ltd¹ · Supply Agreement with Meridian Rail Systems Ltd (draft 3, received 20 October)² · EV⁸

The draft is subject to the NDA dated 2 September 2026. Meridian’s tender for the Western Sydney rolling stock contract has not been announced; our client’s involvement is confidential.²

Commercial terms. Five-year supply of fabricated bogie frames at $18,400 per unit, minimum 400 units per year, with a 3% annual escalation. The client’s cost is approximately $13,900 per unit at current steel prices.³

Issues.

Clause 14, indemnity. Uncapped indemnity for any loss arising from a defect, including consequential loss and Meridian’s liability to Sydney Trains. On our client’s margins, a single derailment-related claim could exceed the contract value.⁴ Recommend cap at 12 months’ fees and exclusion of consequential loss.
Clause 9, steel price. No pass-through of steel price increases. Our client’s supplier, BlueScope, has given notice of an 11% increase from January.⁵ Recommend a rise-and-fall clause tied to the published index.
Clause 22, exclusivity. Our client may not supply bogie frames to any other rail operator in Australia for the term. The client is in early discussions with Queensland Rail and would lose that opportunity.⁶

Fallback positions. Client will accept an indemnity cap at 24 months’ fees if consequential loss is excluded. Client will accept exclusivity limited to New South Wales. Client will not accept the minimum volume without a take-or-pay at 80%, and has told me privately that at the draft price the deal is marginal and they would walk at anything below $17,600 per unit.⁷

Eleanor Vance⁸

An invented memo. Hover or tap any highlighted item for the analysis; the full list is below.

What leaves the firm

The client

Named. The memo is now about an identifiable company. Rule 9 confidentiality guide.

Frontier Models (ChatGPT, Claude)	Transmitted to the provider, generally offshore on personal plans
Ermos Edge	Stays on your instance in Australia; leaves only inside a permitted excerpt, per question, to the named provider
Ermos Dominion	Stays on the unit in your office

The counterparty, the deal and the NDA

That a named company is negotiating with a named counterparty for a named tender, all of which is confidential under an NDA the client signed. Pasting the memo breaches the NDA on the client’s behalf. Rule 9 confidentiality guide.

Frontier Models (ChatGPT, Claude)	An NDA breached on the client’s behalf
Ermos Edge	Stays on your instance in Australia; leaves only inside a permitted excerpt, per question, to the named provider
Ermos Dominion	Stays on the unit in your office

Price, volume and the client’s cost

The commercial heart of the deal, and the client’s margin, which the counterparty does not know. Confidential under rule 9 and commercially the most damaging line on the page. Rule 9 confidentiality guide.

Frontier Models (ChatGPT, Claude)	The client’s margin now on a third party’s servers
Ermos Edge	Stays on your instance in Australia; leaves only inside a permitted excerpt, per question, to the named provider
Ermos Dominion	Stays on the unit in your office

The risk the firm has found

An uncapped indemnity and the firm’s view of its consequence. Privileged advice. Legal privilege guide.

Frontier Models (ChatGPT, Claude)	Privileged advice transmitted; on consumer tiers it can train future models
Ermos Edge	Stays on your instance in Australia; leaves only inside a permitted excerpt, per question, to the named provider
Ermos Dominion	Stays on the unit in your office

A supplier and a price rise

A third party’s name and its pricing notice. Confidential to the client, and to the supplier. Rule 9 confidentiality guide.

Frontier Models (ChatGPT, Claude)	A supplier’s pricing transmitted
Ermos Edge	Stays on your instance in Australia; leaves only inside a permitted excerpt, per question, to the named provider
Ermos Dominion	Stays on the unit in your office

A separate opportunity

That the client is talking to another rail operator. A second confidential negotiation, disclosed inside the first. Rule 9 confidentiality guide.

Frontier Models (ChatGPT, Claude)	A second confidential negotiation transmitted
Ermos Edge	Stays on your instance in Australia; leaves only inside a permitted excerpt, per question, to the named provider
Ermos Dominion	Stays on the unit in your office

The fallback positions and the walk-away price

What the client will accept, what it will refuse, and the number below which it leaves. If the counterparty had this paragraph, the negotiation would be over. Privileged, and the reason review memos never go into external tools. Legal privilege guide.

Frontier Models (ChatGPT, Claude)	The walk-away price transmitted before negotiation begins
Ermos Edge	Stays on your instance in Australia; leaves only inside a permitted excerpt, per question, to the named provider
Ermos Dominion	Stays on the unit in your office

The partner

Ties the memo to a solicitor bound by rule 9. Law Society AI guidance.

Frontier Models (ChatGPT, Claude)	Transmitted, and the leak is traceable to you
Ermos Edge	Stays on your instance in Australia; leaves only inside a permitted excerpt, per question, to the named provider
Ermos Dominion	Stays on the unit in your office

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 Edge	Ermos Dominion
Where the document is processed	Provider servers, generally offshore on personal plans	Your instance in an Australian data centre; answers generated in Australia	The unit in your office
Who can read it	Provider staff and systems, under policy	Your staff, within the boundaries you set. The named Australian provider processes only permitted excerpts, under contract, and keeps nothing	Your staff only
Is it kept	On the provider’s terms	Kept on your instance; the provider is barred from retaining it	Stays on the unit
Can it train a model	On consumer tiers, yes	No, by contract	Only your own unit learns from it, on your hardware

Part of AI data leakage in law firms. See how this works in practice: Contract review on Ermos.

Common questions

Do ChatGPT and Claude use the information in a contract review memo?

Yes, a pasted memo is sent to the AI company's servers to be processed. 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 review memo holds the counterparty, the client's margin and the fallback positions.

Can we use ChatGPT or Claude for a contract review memo if we remove the company names?

No, because the deal is still recognisable once the company names go. 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. The per-frame rate, the annual minimum volume, the clause numbers and the unannounced tender describe one transaction. The client's margin and walk-away rate are damaging under any name. 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.

Can ChatGPT or Claude be trusted to draft a contract review memo?

Not without a solicitor checking every line against the agreement. Language models can write fluent text that contains details missing from the source, wrong figures, wrong dates and references that look real but are not. The costliest errors here would be a wrong clause number, a misquoted indemnity cap, a misstated per-frame rate or a fallback position the client never gave. The reviewing solicitor should check each one against the draft agreement and the client's instructions before the memo goes anywhere.

How can a law firm use AI for a contract review memo 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 solicitor asks for the review in plain language, and the memo is built from the draft agreement and the firm's precedent bank. Each flag cites its excerpt, and the solicitor checks every flag before signing the memo.

Can AI review a contract without the contract leaving the firm?

On Ermos, yes. The draft is indexed on the unit in your office and reviewed against the firm's own precedent bank: unusual clauses, missing protections, defined-term inconsistencies, each flagged with the excerpt it came from. The NDA is honoured because the draft never leaves.

Does the NDA actually cover pasting into an AI tool?

Most NDAs restrict disclosure to anyone other than the receiving party's advisers who need to know. A frontier model provider is a third party under any ordinary reading, and it is the client who signed, with the firm acting on its behalf. The Law Society guidance expects a full contract review before any client information goes into an AI tool, which includes the client's own NDAs.

Why keep the fallback positions in the same memo?

Because that is how review memos are written: issues, recommendations, and what the client will live with. It is good practice inside the firm. It is the reason the whole memo must stay there.

See Dominion answering from your own documents.

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

Book a demo

Keep reading

AI data leakage in law firms AI for law firms The Law Society AI guidance All professions