Review submissions and AI confidentiality: what a Tribunal statement after a visa refusal gives away in ChatGPT or Claude.

After a refusal, the migration agent prepares a statement of facts, issues and contentions for the Administrative Review Tribunal. It sets out the decision, the evidence and the argument for why the decision should change, and it is the document where staff are most tempted to use ChatGPT or Claude for drafting and structure. The Tribunal published a practice direction on generative AI in August 2026, and it makes the person who prepares a submission responsible for checking it. This page marks every item that the Code of Conduct, the Tribunal’s practice direction and the Privacy Act bear on, and shows how the same draft is produced on Ermos inside the practice.

The review submission after a refusal

Prepared after a visa refusal and filed with the Administrative Review Tribunal in support of an application for review. Section 276 of the Migration Act treats preparing for and representing a person in proceedings before the ART as immigration assistance. The Tribunal’s Generative AI Practice Direction 2026 applies to all persons involved in proceedings.

Administrative Review Tribunal
Statement of facts, issues and contentions · Tendai Moyo and the Minister, ART case 2026/00000 (placeholder)¹ · 29 October 2026

AI disclosure. A generative AI tool (being ChatGPT) was used to check the chronology below and to improve grammar. The statement has been reviewed and checked by Renata Kowalczyk-Bell.⁸

Decision under review. On 9 October 2026 a delegate refused the applicant’s subclass 190 visa, saying they were not satisfied that she had five years of skilled employment as a civil engineering technician. The delegate found her claimed 15 employment points were not established.²

Facts. The applicant worked for Marlow Bridge Consulting Pty Ltd in Sydney from 4 February 2019 to 27 March 2026 as a civil engineering technician, full time, on a salary of $92,000 at the end. She supervised survey crews and prepared drawings.³ The employer’s director, Mr Kessler, withdrew a reference letter after a dispute about the applicant’s resignation, and the applicant says he has told others she left without notice.⁴

Evidence. Payslips for each year, ATO income statements for 2019 to 2025, and bank statements for account BSB 000-000 ending 4417 (placeholder) showing fortnightly salary credits.⁵

Contentions. The applicant’s total is 70 points. If the employment points fell away it would be 55, below the threshold. The payslips, income statements and bank deposits show continuous full-time employment, and the delegate gave no reason to doubt them.⁶

Personal circumstances. The applicant lives in Liverpool with her husband, Brian, and their two children. She is on a bridging visa and has been told she must leave Australia if the review fails.⁷

Renata Kowalczyk-Bell, registered migration agent, MARN 0000000, Kingfisher Lane Migration, representative for the applicant.⁹

An invented Tribunal statement. The people, the employer, the case number and the account are invented or placeholders. The points total is the invented practice’s working figure. Hover or tap any highlighted item for the analysis; the full list is below.

What leaves the practice

The applicant and the case

A named applicant, the respondent and a case number. Personal information, and the key that ties the statement to a live proceeding. Privacy Act guide.

Frontier Models (ChatGPT, Claude)	An applicant and a live case 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 decision, the date and the delegate’s reasons

The refusal and why. Information from the applicant’s own immigration record, and the starting point for the argument. Privacy Act guide.

Frontier Models (ChatGPT, Claude)	A refusal and the delegate’s reasons 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

Employment facts

A named employer, dates, a role and a salary. Personal and employment information the applicant is relying on to prove a points claim. Privacy Act guide.

Frontier Models (ChatGPT, Claude)	Employer, dates, role and salary transmitted; on consumer tiers they 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 dispute with a former employer

A named director, a withdrawn reference and an allegation about what he has said to others. Sensitive for the applicant and the employer, and the kind of statement a person may not want repeated anywhere. Privacy Act guide.

Frontier Models (ChatGPT, Claude)	An allegation about a named director 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

Evidence and a partial account number

The documents relied on and part of a bank account. Financial information, and the evidence a Tribunal member will weigh. Privacy Act guide.

Frontier Models (ChatGPT, Claude)	Evidence and part of a bank account 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 argument and the points arithmetic

Why the delegate was wrong and the stakes in points. The practice’s work product. The Tribunal’s practice direction says the person who prepares a submission must check facts, authorities, evidence and chronologies, so the total and the claims need checking by a person.

Frontier Models (ChatGPT, Claude)	The argument transmitted, and the model can still invent a citation or misstate the evidence
Ermos Edge	Stays on your instance, but a model can still misstate the evidence, so the agent checks every claim
Ermos Dominion	Stays on the unit, but a model can still misstate the evidence, so the agent checks every claim

Personal circumstances

Where the family lives, the husband, two children and the consequence of failure. Personal information about people who are not the applicant. Privacy Act guide.

Frontier Models (ChatGPT, Claude)	A family’s circumstances and a risk of removal 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 AI disclosure

The Tribunal’s practice direction says disclosure should be prominent, should say what tool was used, where and how the output was checked, and can be brief. This sentence is the practice’s own record of how the draft was made.

Frontier Models (ChatGPT, Claude)	Describes how the draft was made; the sentence itself holds no client detail
Ermos Edge	Nothing to keep in; the disclosure is the practice’s own duty
Ermos Dominion	Nothing to keep in; the disclosure is the practice’s own duty

The representative

The agent’s name and MARN. The Tribunal can refer a representative’s conduct to a regulator when it may involve a breach of a professional obligation.

Frontier Models (ChatGPT, Claude)	The agent’s name and MARN are published to attract clients; low exposure on its own
Ermos Edge	Nothing to keep in; these are business details
Ermos Dominion	Nothing to keep in; these are business details

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 practice 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 migration practices.

Common questions

Do ChatGPT and Claude use the information in an ART review submission?

Yes, whatever is pasted from the statement 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. The statement holds a live case, the delegate's reasons, a dispute with a named director and part of a bank account number.

Is it safe to paste a Tribunal review submission into ChatGPT or Claude?

Not without meeting what the Tribunal requires. The ART's Generative AI Practice Direction says a person must not enter confidential, private, privileged or otherwise protected information into a publicly available tool unless satisfied that doing so is lawful, authorised and consistent with every obligation that applies. 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. This statement names the applicant, her employer, her children and her bridging visa.

Can ChatGPT or Claude be trusted to draft a Tribunal review submission?

No. Language models can produce fluent text with details that are not in the source, wrong figures, wrong dates or references that look real but are not. A qualified person should check every item against the file. In this statement the points arithmetic, the employment dates and the account of the delegate's reasons would do the most harm if wrong.

How can a migration practice use AI for an ART review submission 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 statement is drafted from the client file and the practice's own template. The registered migration agent checks every fact and contention against the evidence, signs it, and can explain how it was checked, as the practice direction expects.

What does the Tribunal expect of a representative who uses generative AI?

The Administrative Review Tribunal's practice direction, signed on 20 August 2026, applies to everyone involved in proceedings. The person responsible for a document must check that facts, authorities, evidence and chronologies are accurate and traceable. The Tribunal may require disclosure of AI use, and a person must be able to explain what was used and how the output was checked.

Can confidential material go into a publicly available AI tool for a Tribunal matter?

The practice direction says a person must not enter confidential, private, privileged or otherwise protected information into a publicly available generative AI tool unless satisfied that doing so is lawful, authorised and consistent with every obligation that applies. It also warns that restricted material can be an issue even with a ringfenced tool.

What happens if an AI tool produces something wrong in a submission?

The Tribunal may ask for the material to be corrected or withdrawn, ask how AI was used, give the material less weight or decline to rely on it. If a representative's conduct may breach a professional obligation, the Tribunal may refer the matter to a regulator, which for an agent is the OMARA.

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 migration practices Resources The Privacy Act, the APPs and AI All professions