Australia to review copyright law for AI training: what legal teams need to know
Australia will convene a two-day meeting beginning Monday to reassess copyright settings in light of AI training practices. Officials have already ruled out a free text-and-data-mining (TDM) exception, signaling a shift toward paid access and stronger enforcement.
The Attorney-General's Department said the session will be led by the Copyright and AI Reference Group, bringing in voices from tech and the creative sectors. The goal: protect creators while keeping room for innovation.
What's on the table
- No TDM exception for AI training. AI developers won't get free use of protected works.
- Exploring a licensing pathway under the Copyright Act so model training triggers payment to rightsholders.
- Lowering cost and friction of enforcement to help creators act against unlicensed use.
- No relaxation of existing copyright rules. As stated, the plan is to balance incentives for creators and developers, not weaken protection.
Why this matters for in-house and outside counsel
- Anticipate licensing negotiations: model training, dataset ingestion, embeddings, and derivative use will need explicit terms and audit rights.
- Data governance: advise AI clients to inventory training sources, document provenance, and segregate high-risk content.
- Enforcement posture: expect faster, cheaper remedies; prepare claim templates, evidence capture workflows, and API-level monitoring.
- Territoriality: clarify whether Australian-sourced data triggers obligations for offshore training runs and cross-border data flows.
- Contracts: update supplier and research agreements to cover TDM, caching, model outputs, indemnities, and sunset/retention duties.
The fault lines
Creators say current law didn't anticipate model training on massive, often protected datasets-and they want compensation baked in. Tech stakeholders warn that overly tight rules will slow progress and put Australia behind peers.
"AI systems rely on large volumes of data, much of which is created by human effort and creativity," said Michelle Rowland. She added that the tech and creative industries need a compromise that works for both sides.
How Australia's stance compares
Countries like Japan and Singapore allow limited TDM exceptions. Australia's decision to withhold a broad exception moves it closer to a paid-use framework. If a licensing mechanism lands, expect standardized rates, collective management discussions, and compliance audits to follow.
For context on current policy settings, see the Attorney-General's Department overview of copyright policy here.
Action items before the meeting wraps
- Rightsholders: compile registries of works, confirm ownership chains, and flag high-value catalogs for priority licensing.
- AI developers: pause unlicensed ingestion of protected corpora; document consent, licenses, and opt-outs for existing datasets.
- Both sides: draft model clauses for training rights, data lineage, output restrictions, and compliance audits.
- Prepare submissions: be ready to engage the Reference Group with pragmatic proposals on rates, safe harbors, and dispute resolution.
What to watch next
- Whether the Copyright Act will expressly recognize training-use licenses and collective administration.
- Procedural reforms that cut costs for takedowns, discovery, and injunctions against unlicensed use.
- Any carve-outs for research or public-interest uses-and how "publicly available" will be treated.
If your team needs a fast primer on practical AI concepts for policy, contracts, and risk, browse AI courses by job here.
Your membership also unlocks:
