Shortlist spreadsheets and AI data leakage: what a candidate comparison sheet gives away in ChatGPT or Claude.

Recruiters compare candidates in a spreadsheet because it sorts, filters and goes to the client as an attachment. Someone then uploads the sheet to ChatGPT or Claude to rank the field, write the cover note or tidy the comments. The sheet puts seven people side by side: what each earns and wants, how long they must stay, their visa status, a rating, and a remarks column where consultants write things that were never meant for a client. This page marks what a routine shortlist exposes under the Privacy Act and anti-discrimination law, and shows how the same sheet is worked on Ermos inside the agency.

The shortlist spreadsheet

Built by the consultant when submitting candidates for a role, kept for the life of the search and often emailed to the client in a cleaned-up form. The version with the remarks column stays internal, which is the version that gets pasted into an AI tool.

fx
ABCDEFG
1CandidateCurrentExpectsNoticeWork rightsRatingNote
2
3
4Needs sponsorship transferred.
5
6Interviewing elsewhere this week.
7
8Light on month-end.
Northgate FM shortlist

An invented spreadsheet. All candidates are fictional. Hover or tap any highlighted item for the analysis; the full list is below.

What leaves the agency

  1. 1
    Candidate names

    Seven named people on one sheet, each of whom applied in confidence. Personal information held by the agency, and the key that ties every other column to a real person. Privacy Act guide.

  2. 2
    Current pay and expected pay

    What each person earns and what they want. Personal information, and the figure a client uses to anchor an offer. It is the information that, in a recruiter’s hands, is meant to go to one client at a time. Privacy Act guide.

  3. 3
    Notice periods

    How soon each person can leave. It also shows each of them is looking, which their current employers do not know.

  4. 4
    Visa and work rights

    A temporary visa and its expiry date for one candidate. Personal information, and information that can point to national origin. The Privacy Act list of sensitive information includes racial or ethnic origin, but not immigration status as such. Privacy Act guide.

  5. 5
    The agency’s ratings

    The consultant’s assessment of each person. Personal information in the form of an opinion, and the agency’s own work product. If an AI tool re-scores the field from this column, the OAIC’s AI guidance raises accuracy and human oversight. Privacy Act guide.

  6. 6
    Remarks about age, family and caring

    A candidate’s age set beside a client’s stated preference, a parental leave return and a caring break. The Fair Work Act stops a prospective employer taking adverse action against a prospective employee because of attributes that include age and family or carer’s responsibilities, and the NSW Act has provisions aimed at employment agencies. A written remark like this is the kind of record a complaint turns on. Privacy Act guide.

  7. 7
    A conviction note

    A remark about a conviction the candidate says is spent. Criminal record is sensitive information under the Privacy Act. Under the Criminal Records Act 1991 (NSW) a person generally need not disclose a spent conviction, and the Australian Human Rights Commission can look at complaints of discrimination on the ground of irrelevant criminal record. Privacy Act guide.

  8. 8
    The rating formula

    A test of a rating against a threshold, with no candidate data in the formula itself. The same formula can be shared safely; the sheet it sits in cannot.

  9. 9
    Other tabs, hidden columns and file properties

    A spreadsheet file generally carries more than the visible tab: other sheets such as the candidates not submitted, hidden columns, cell comments and properties such as the author. Uploading the file sends all of it, including what is off screen.

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 agency 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 candidate 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 AI data leakage in recruitment agencies.

Common questions

Do ChatGPT and Claude use the information in a shortlist spreadsheet?
Yes, everything pasted or uploaded 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 sheet sets seven candidates' current and expected pay beside notice periods, visa status, ratings and a remarks column about age, family and a conviction.
Can we use ChatGPT or Claude for a shortlist spreadsheet if we remove the candidates' names?
No, because the other columns still describe real people. 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. A 482 visa to 14 August 2027, a candidate aged 52 and a note about a 2012 conviction each point to one person on a sheet of seven. 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 I upload a shortlist spreadsheet to ChatGPT or Claude instead of pasting it?
Uploading sends the whole file out of the agency. A spreadsheet can carry other sheets, hidden columns, comments, tracked changes and file properties as well as the tab you meant to share. The OAIC's guidance also tells organisations to review carefully the terms and settings that will apply to their use of any AI product. In this workbook a Not submitted tab holds the candidates held back, and the internal remarks column sits beside the visible ones.
How can a recruitment agency use AI for a shortlist spreadsheet 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. Ermos works from the candidate files and the agency's own shortlist template. The consultant checks every rating and remark and signs off the sheet before any version goes to a client.
Is a candidate's visa status sensitive information?
The Privacy Act list of sensitive information covers racial or ethnic origin, health, criminal record, religious beliefs and similar categories, and does not name immigration status. A visa and its expiry date are still personal information, and they can point to national origin, so they belong in the file only where the role needs them.
Can a client's preference for a younger candidate go on our sheet?
Age is one of the attributes the Fair Work Act protects when a prospective employer refuses to employ or offers different terms, and, where a complainant alleges a prohibited reason, section 361 presumes that was the reason unless the person who took the action proves otherwise. The NSW Act also has provisions aimed at employment agencies. A written age remark is evidence, so take advice before recording or acting on one.
What does the Privacy Act require us to do about a sheet like this?
APP 11.1 requires an APP entity to take reasonable steps to protect personal information from misuse, interference and loss and from unauthorised access, modification or disclosure. A sheet of seven candidates' pay and notes uploaded to a tool the agency does not control is the kind of step a reasonable agency has to be able to justify.

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 recruitment agencies Resources The Privacy Act, the APPs and AI All professions