AI data leakage in recruitment: what leaves when candidate files go into ChatGPT or Claude.

A recruitment agency holds the working lives of hundreds of people: current salaries, why they want to leave, the health and family circumstances they mention in a screening call, visa status, referees’ candid opinions and sometimes a criminal record. The Privacy Act’s employee records exemption does not cover prospective employees, and the OAIC says it is unlikely to apply to organisations that provide recruitment services under contract to an employer. Discrimination law reaches hiring decisions and the notes behind them. This series takes ten documents an agency produces every week, from the CV to the rejection email, marks every item those obligations touch, and shows where each one goes when the document is pasted into ChatGPT or Claude. Each page ends with the same document on Ermos.

The documents, one at a time

RecordThe candidate CV and profile

Address, birth date, salary, a career break, a parish role, referees and what the consultant really thinks. One person, one document.

See what leaves → SpreadsheetThe shortlist spreadsheet

Seven candidates side by side: pay, notice, visa, rating and the remarks column nobody was meant to read.

See what leaves → NotesThe reference check notes

A warning from 2019, a redundancy, a health remark, a pay figure and the thing the referee asked you not to repeat.

See what leaves → FormThe client job order form

A predecessor who left, an unannounced restructure, a hidden budget ceiling and a preference no agency should write down.

See what leaves → LetterThe offer letter

The package, a sign-on payment that reveals her current bonus, a police check condition and a note about the agency’s fee.

See what leaves → TranscriptThe candidate interview transcript

Everything a candidate says when she thinks she is talking to one person. Salary, health, family plans and a rival offer.

See what leaves → SpreadsheetThe salary benchmark spreadsheet

Market bands beside named employees’ real salaries, the agency’s fee per role and a note about who might leave.

See what leaves → AgreementThe terms of business

The fee percentage, the guarantee, the introduction period and the clause that says who holds a licence.

See what leaves → EmailThe candidate rejection email

The reason the client gave, feedback from the interview, another candidate’s salary and a note to self left in the draft.

See what leaves → Text or chat messageThe candidate chat messages

A pregnancy, a job search kept secret, a pay figure, the client’s budget and a father’s hospital visit. All in a text thread.

See what leaves →

How the leakage happens

The mechanism is the same for every document. The whole text 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. On consumer tiers it can be retained and used to train future models. The provider’s staff and systems can reach it under the provider’s own policies. Once sent, the agency no longer controls it. Our on-premise vs cloud comparison sets out the four ways to run AI and where each one puts your documents.

The obligations in play

Each marked item on these pages links to the guide that covers its obligation:

The same documents on Ermos

Ermos Dominion drafts and searches on a unit in your office, so the documents never leave it. Ermos Edge keeps your documents on your own instance in an Australian data centre and sends only permitted excerpts, per question, to a named Australian provider. Every page in this series ends with both, item by item.

Common questions

Can recruiters use ChatGPT or Claude with candidate and client information?
That depends on the agency's position under the Privacy Act, which can still apply to candidate files. 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 OAIC says the employee records exemption does not cover personal information about prospective employees such as unsuccessful job applicants, and is unlikely to apply to organisations that provide recruitment services under contract to an employer. A small business operator can sit outside the Act, but under section 6D(4)(c) it loses that status if it discloses personal information about another individual for a benefit, service or advantage, unless the individual consents or the law requires or authorises it.
Do ChatGPT and Claude train on what staff paste in?
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. For a recruitment agency the question matters because what staff paste is a candidate's salary, health, family circumstances and secret job search.
Do business plans such as ChatGPT Business or Claude Team solve the problem for recruiters?
Only in part. A business plan changes the default on training, but it does not change who receives the content. 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. The OAIC's guidance also tells organisations to review carefully the terms and settings that will apply to their use of any AI product. For an agency, a shortlist or a set of reference notes would still reach the AI company in full. Our comparisons with ChatGPT Enterprise and Claude weigh the trade-offs.
What is the safest way for a recruitment agency to use AI with candidate documents?
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. For recruiters, staying in control means the candidate file stays where the agency can see it, correct it and answer for it.
What is AI data leakage in a recruitment agency?
Candidate or client information leaving the agency's control through an AI tool. When a CV, a shortlist or a set of reference notes is pasted into ChatGPT or Claude, it is transmitted to the provider's servers and can be retained on the provider's terms. For a recruiter the leaked information is a person's salary, health, family and job search, and the agency's own client relationships.
Does the Privacy Act apply to a small recruitment agency?
Not automatically. A small business operator (turnover of $3 million or less) is outside the Act, but under section 6D(4)(c) an operator that discloses personal information about another individual to anyone else for a benefit, service or advantage loses that status unless the individual consents or the disclosure is required or authorised by law. Whether an agency is covered depends on its facts, so check your own position.
Is there guidance on using AI in recruitment in Australia?
The OAIC's guidance on commercially available AI products (October 2024, updated January 2025) recommends not entering personal information, particularly sensitive information, into publicly available generative AI tools, and discusses AI screening of candidates as an accuracy risk. The Australian Human Rights Commission lists an AI and recruitment compliance checklist. From 10 December 2026, APP 1.7 adds privacy policy disclosure for APP entities that use a computer program to make, or substantially support, decisions that could significantly affect an individual.

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
Resources The Privacy Act, the APPs and AI All professions