The record
When old industrial land is redeveloped, an environmental consultant investigates it and writes a chain of reports: a study of the site's history, a detailed investigation that tests soil and groundwater against health limits, a plan for the clean-up, and a final report with test results from after the clean-up.
In Victoria and New South Wales a second expert, accredited by the state's environment regulator, then reviews that work and signs a formal statement on whether the land is fit for its proposed use. In Victoria a lighter first step screens out sites that do not need a full audit.
The record is the consultant's reports with their laboratory tables, the auditor's review and reasons, and a ruling: fit, fit with conditions, or not fit.
A real example, from a statement the New South Wales regulator hosts in public. The site is one lot of a former lead and zinc smelter near Newcastle. Several consultancies had investigated and cleaned it up over the years, and earlier audit statements for the smelter land go back to 2004 1. A site auditor at another firm reviewed the reports and issued a statement with an audit report dated 2018 1. The statement records the auditor's conclusion on fitness for the stated use, with any conditions, such as a plan for continuing management.
The numbers are on the Victorian side. The regulator's register there was queried on 12 October 2026 and returned 5,451 2 audit records. Nearly all of these carried a link to documents, and most a link to appendices. No individual Victorian report was opened, so what the appendices hold is inferred.
Why it might matter for training AI
Vendors' pages show labs asking for expert judgement with reasons and for private tests. Here one accountable expert rules on another's long technical work, across several documents, with laboratory figures that give a second, numerical check. No vendor's or lab's page names environmental work; the fit is inferred.
No purchase of this record by a lab or a vendor was found.
Who holds it and who may share it
Victoria publishes the reports. The regulator's public register lists screening and audit reports submitted since 1991, each with the report attached 3. Its guidance says the auditor must give the statement and report to the regulator within 5 4 business days of finishing, and to the client and the planning authority. The register shows completed audits running at a couple of hundred a year 2.
Public to read is not the same as free to reuse. The register's index is an open government dataset under a Creative Commons attribution licence 5. The regulator's copyright page licenses its own material the same way but excludes content supplied by third parties, and warns that their permission may be needed 6. The audit reports are written by private auditors and consultants.
New South Wales receives statements, not usually reports. The regulator's page says that for audits required by law the auditor must send the statement to it and to the local council 7. Full reports reach the public mostly as attachments to planning applications. That may change: minutes of a 2025 meeting between the regulator and auditors record a proposal that auditors routinely lodge full reports as well 8. Whether it was adopted was not established.
Whether consultants and auditors keep copyright in their reports was not checked against any firm's terms.
Who is already there
Several American software companies draft or read environmental site reports for consultancies and lenders. One states that customer data never trains its models 9. All work on one customer's files at a time. None was found holding reports from many firms, a public test, or anything that pairs a consultant's work with an independent auditor's ruling.
The New South Wales regulator told auditors in 2025 that it had no formal policy on AI and would form a working group 8.
No dataset, test or paper using the Victorian register was found.
What would have to be stripped
Little. The reports name owners, auditors and consultants, and now and then neighbours and home addresses. The site's address is part of the record and already public.
What stands in the way
- No buyer has been identified.
- The report texts belong to third parties. Reading public reports is one thing; copying thousands into something sold for training is another. Whether Australian copyright law has an exception for this was not established.
- The outcomes vary little. The public record shows the signed-off state, and sites that fail tend to be cleaned further until they pass. The telling content is the conditions and the reasons, not the yes or no.
- The reports are long scanned documents with laboratory tables, so extracting them is real work.
- The richest material, the auditor's comments on the consultant's drafts, is on no register.