OpenAI and Anthropic seek a narrow Australian AI-training copyright route
The companies urged a parliamentary inquiry to reconsider restrictions on training with creative material. Consent, payment and the shape of any conditional approval remain contested.
The short version
- What happened
- Reuters reports that both companies asked Australia to consider a limited route for model training under conditions.
- Who it affects
- Authors, artists and publishers need clarity about consent, payment and remedies before their work is used in commercial training.
- What remains unproven
- This account relies on Reuters' reporting of inquiry submissions because the original parliamentary materials were not accessible in our verification path.
Keep in mind
This account relies on Reuters' reporting of inquiry submissions because the original parliamentary materials were not accessible in our verification path.
The inquiry has not reported and no detailed approval or payment scheme is settled.
Source analysis · 1 linked source. Primary analysed source date: 22 September 2026.
Read the full report · 3 min The evidence, methods, implications and limitations
The proposals before the inquiry
OpenAI and Anthropic submitted positions to an Australian parliamentary inquiry on AI laws in September, according to Reuters. Anthropic acknowledged that broad copyright exceptions had not gained support and proposed a narrower form of conditional approval for training. OpenAI called for a framework that permits learning from publicly available information while offering collaboration with rightsholders. These are the companies' preferred policy routes, not the government's adopted position.
Reuters reports that Australia has already ruled out an exemption from copyright law that would facilitate the companies' local training activity. The firms linked their proposals to infrastructure investment in the country. Investment may be relevant to public policy, but it does not by itself answer whether authors, publishers, musicians and artists have consented or how any value from their work should be shared.[1]
The unresolved bargain
A conditional approval could mean different things: licensing, disclosure, payment, limits on data categories, an opt-out mechanism or independent audit. The reporting does not establish which conditions lawmakers would accept, what rights holders would receive or how a creator could check whether a work was included. Calling the proposal a compromise before those details are written would be premature.
The same framework must account for both large archives and individual creators whose work is widely accessible online but not freely licensed for commercial training. Developers seek predictable rules for assembling datasets; creators seek bargaining power and a remedy when their rights are ignored. A useful process should publish how the rules are enforced, not just how they are described.[1]
What a decision would need to show
The committee's report, due in November, is the next public checkpoint. Its recommendations may differ from either company's submission, and any change to copyright law would require a separate governmental and legislative process. The inquiry should make visible the expected economic benefit, distribution of payments, rights of refusal and practical oversight costs.
Other jurisdictions are watching the same tension between model development and creative rights. Australia's choice will matter beyond one data-centre deal if it creates a workable template for consent and compensation. For now the defensible description is a contested proposal within an ongoing inquiry.[1]