News analysis · 22 September 2026
Australia’s AI Copyright Debate: Build a Training-Rights Register
By the ELYMENT AI editorial team · Free to read
OpenAI and Anthropic have asked Australia to reconsider how copyright law applies to AI model training, Reuters reported on 22 September 2026. Their submissions proposed narrower pathways after the government ruled out a broad text-and-data-mining exception. No new exemption has been enacted. For businesses, the immediate task is not to predict the policy result. It is to maintain a training-rights register that proves what data enters each AI process, who controls it, which uses are permitted and how material can be removed.
What changed in Australia’s AI copyright debate
Reuters reported that Anthropic proposed a narrow form of conditional approval for AI training and said it was open to conditions supporting Australian creators and cultural endeavours. OpenAI called for a balanced framework that would allow models to learn from publicly available information while giving rightsholders opportunities to collaborate. Both companies connected copyright settings to planned Australian infrastructure investment.
These are submissions to a parliamentary inquiry, not enacted rules. Reuters said the Joint Select Committee on Artificial Intelligence is due to report in November 2026. The proposals therefore belong in a policy watchlist, not a production permissions table. A company cannot treat a vendor’s preferred policy as authority to use a work.
Consultation is not permission
Australia’s Attorney-General’s Department says the government is not considering a text-and-data-mining exception. Its Copyright and Artificial Intelligence Reference Group is instead examining fair, legal licensing, certainty for AI-generated material and lower-cost enforcement. The department also identifies AI inputs and copyright-related transparency as priority issues.
ABC reported on 16 September that the government was consulting creators, media organisations and AI companies about possible models intended to provide control and payment. Options described in leaked material drew criticism from creators and political opponents. That contested consultation does not change the operational baseline: teams should verify the rights basis for each intended use and obtain legal advice where it is uncertain.
Build one training-rights register
Keep a decision record for every dataset, feed or collection used to train, fine-tune or evaluate a model. Retrieval and prompt context should be recorded separately because access for one purpose does not establish permission for another. At minimum, capture:
Make the register a release gate. If the rights basis, purpose or removal path cannot be evidenced, exclude the material until the gap is resolved rather than allowing policy uncertainty to become undocumented production exposure.
- the exact source, version, acquisition date, content classes and accountable owner;
- the rights basis, licensor, territory, term, permitted users and permitted AI purposes;
- provenance evidence showing how material entered the collection and whether contributors had authority;
- restrictions covering model training, fine-tuning, evaluation, retrieval, outputs and commercial use;
- opt-out, correction, deletion and downstream-removal procedures with tested response times;
- vendor warranties, audit evidence, indemnities and notice duties for third-party datasets; and
- the approving legal or risk owner, unresolved assumptions, expiry date and next review trigger.
What business leaders should do now
Ask product, data, procurement and legal teams to sample the three highest-value AI workflows. Trace every training, fine-tuning and evaluation source to a documented rights basis. Separate public availability from permitted use, and verify that supplier contracts match the actual workflow, territory and commercial purpose.
Then add a policy-change trigger. When Parliament reports or government guidance changes, update the affected register entries and controls rather than rewriting the whole AI programme from memory. The same evidence also supports vendor diligence, creator enquiries, incident response and model retirement.
ELYMENT AI helps organisations convert uncertain AI policy into controlled operating decisions. Start with one training-rights register that connects data provenance, permission, purpose, removal and accountable approval before expanding the model or workflow.
Sources
- Reuters, Anthropic and OpenAI call for Australia to revisit AI training rules (22 September 2026) - Independent reporting on the companies’ inquiry submissions, their proposed approaches, investment context and the committee timetable.
- Attorney-General’s Department, Copyright and Artificial Intelligence Reference Group (Accessed 22 September 2026) - Primary government source on Australia’s AI-copyright consultation priorities, licensing work and stated position on a text-and-data-mining exception.
- ABC News, AI investment meetings and copyright consultations (16 September 2026) - Independent Australian reporting on the consultations, reported options, government comments and responses from creators and political representatives.
Continue learning
Frequently asked questions
Has Australia created an AI training copyright exemption?
No. The reported OpenAI and Anthropic positions are inquiry submissions, while the government has said it is not considering a broad text-and-data-mining exception.
Does publicly available content automatically permit AI training?
No. Public access does not by itself establish the copyright, contract, privacy or permitted-purpose basis for training. The intended use and applicable rights still need verification.
What is a training-rights register?
It is an auditable record linking each AI dataset or source to provenance, rights, permitted purposes, restrictions, removal procedures, supplier evidence and accountable approval.