Claims emails and AI confidentiality: what bad news to a client gives away in ChatGPT or Claude.

The hardest email a broker writes is the one that tells a client their claim may not be covered. It has to be accurate, careful about the brokerage’s own position, and kind, which is the exact brief that gets handed to ChatGPT or Claude. It carries the insurer’s reasoning, the client’s exposure, and often the brokerage’s. This page marks every item in a claims email that the Code of Practice and the Privacy Act bear on, and shows how the same email is drafted on Ermos inside the brokerage.

The claims email to a client

Sent when an insurer raises an indemnity issue or declines a claim, to explain the position and the client’s options, as section 7.1 of the Code of Practice contemplates.

To: Daniel Okafor¹
From: Jenny Nguyen, Claims Manager, Pacific Harbour Insurance Brokers⁸
Subject: Tupou claim¹: insurer’s position and next steps

Dear Daniel,

I wanted to set this out in writing after our call, because the position is serious and you should have it in front of you.

The insurer’s position. Meridian has reserved its rights on indemnity. Its solicitors say that removing the guard from the edgebander may amount to reckless conduct, which the policy excludes. They rely on Marcus Hale’s statement that the guard was taken off two days before the incident because it kept jamming.²

What that means. If Meridian declines indemnity, the claim by Mr Tupou, currently pleaded at $240,000, would fall to the company. Mr Tupou’s surgeon assesses a 12% permanent impairment of the finger and psychological effects, so the figure is unlikely to come down much.³

The workers compensation gap. Separately, the adjuster has noted that your workers compensation policy lapsed between 3 and 14 March for non-payment. Mr Tupou was a labour hire subcontractor, so the lapse may not affect his claim directly, but it is a compliance breach in its own right and SIRA, the NSW workers compensation regulator, may pursue it. We should have flagged the lapse when the renewal invoice went unpaid, and I am sorry we did not.⁴

Your exposure. If the exclusion is upheld and the company cannot meet the claim, there is a question about whether Mr Tupou’s lawyers would look to you personally as the director, particularly given the SafeWork notice. That is a legal question and you need your own advice on it.⁵

What we recommend. Do not give any further statement to the adjuster without legal advice. Engage a solicitor; we can suggest two. Let us respond to Meridian on the indemnity question on your behalf, arguing that the guard removal was by an employee without the company’s knowledge. Keep the current policies paid and current.⁶

I know this comes at a bad time with the Coastline situation. Call me any time.⁷

Jenny⁸

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

What leaves the brokerage

The client and the claim

Named, with the claimant’s surname in the subject line. Privacy Act 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 insurer’s reasoning and a named witness

Why the insurer may decline, and the employee whose statement it relies on. Evidence in a dispute, attributed. Privacy Act guide.

Frontier Models (ChatGPT, Claude)	A named witness and the insurer’s reasoning 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 claimant’s injury and the amount

A third party’s permanent impairment and psychological effects, and the sum claimed. Sensitive information about someone who is not the client. Patient data guide.

Frontier Models (ChatGPT, Claude)	A third party’s injury 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 lapsed policy and the brokerage’s apology

A compliance breach by the client, and an admission that the brokerage should have caught it. The sentence a professional indemnity insurer would read first.

Frontier Models (ChatGPT, Claude)	Your own apology 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 director’s personal exposure

That the client may be personally liable. The most alarming paragraph in the email, about a named individual. Privacy Act guide.

Frontier Models (ChatGPT, Claude)	A director’s personal exposure 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 advice

What the brokerage recommends, including how it will argue the indemnity point. Strategy in a live dispute.

Frontier Models (ChatGPT, Claude)	Strategy in a live dispute 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 reference to the client’s other troubles

The Coastline situation, in passing. Confidential business information.

Frontier Models (ChatGPT, Claude)	Transmitted and kept on the provider’s terms
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 claims manager

Ties the email, and its apology, to a named staff member.

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 brokerage 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 client 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 insurance brokerages. See how this works in practice: Claims chronologies on Ermos.

Common questions

Do ChatGPT and Claude use the information in a claims email to a client?

Both receive every word of the email on the AI company's servers once it is pasted. 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 claims email carries the claimant's permanent impairment, the insurer's reasoning and the director's personal exposure.

Is it safe to paste a claims email to a client into ChatGPT or Claude?

No, because the email is a written record of one dispute and the people in it. The subject line carries the claimant's surname, the body names the witness the insurer relies on, and the figures give the sum claimed. Taking out the surname would still leave a dispute that anyone involved would recognise. 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.

Does deleting the chat remove a claims email to a client from ChatGPT or Claude?

Not at once. Deleting a chat removes it from the user's history, but OpenAI and Anthropic each say they can keep a deleted chat on their systems for up to 30 days, and longer where safety or legal obligations apply (OpenAI, Anthropic). A claims manager can draft a difficult reply and delete the chat the same afternoon. The lapsed workers compensation cover, the claimant's injury and the brokerage's strategy on indemnity would still have gone to the AI company before the chat was deleted.

How can a brokerage use AI for a claims email to a client 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 email is drafted from the claim chronology, the insurer's correspondence and the policy wording, in the brokerage's own tone. The claims manager edits it, checks it and signs it before it is sent.

Why is the brokerage's apology the item to worry about?

Because it is an admission that the brokerage missed a lapse, in writing, and it belongs in exactly one place: the email to the client, where the Code of Practice's commitment to honesty puts it. On a frontier model provider's servers it is the same admission, outside the brokerage's control, retained on terms the brokerage never saw.

Does the Code of Practice cover claims emails?

Section 7.1 commits subscribers to keep clients informed as a claim progresses and, if the insurer declines it, to explain the insurer's reasons and the further steps open to the client. Section 5.1 commits them to timely, plain-English communication and section 3.1 to acting honestly and with integrity. An email like this one is where those commitments are put into practice. Where it was drafted is a separate question the Code does not answer and the Privacy Act does.

How is a difficult claims email drafted on Ermos?

From the claim chronology, the insurer's correspondence and the policy wording on your own unit, in the brokerage's tone. The claims manager edits and sends it. Nothing about the claimant, the insurer's position or the brokerage's own mistake leaves the office to write it.

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 insurance brokerages AI for insurance brokers The Privacy Act, the APPs and AI All professions