Anthropic Told Australia Its Copyright Law Makes AI Training There Impossible

Key points

• Anthropic told an Australian parliamentary inquiry that current copyright law makes it "impossible" to train AI models inside Australia, even as it denied lobbying to change those laws.
• OpenAI apologised a second time before the same committee about a hack on Medicare, Australia's national health insurance system, that occurred last month.
• Media and entertainment groups told the inquiry that weakening copyright protections for AI training would leave artists as, in their words, "roadkill".
• The ABC, Australia's national public broadcaster, told the same committee it doesn't believe existing copyright law needs changing.

Anthropic's message to Canberra this week was, at first glance, contradictory. The company told a parliamentary committee examining AI regulation that it has "never tried to dictate" what Australia does with its copyright law. In almost the same breath, it said those existing rules make training an AI model on Australian soil "impossible."

That's not quite as contradictory as it sounds. Anthropic is distinguishing between lobbying for a specific outcome and simply describing a legal constraint. The practical effect is the same, though: without a change, large-scale AI development cannot happen in Australia under current conditions.

What does copyright have to do with training an AI?

Training a large language model, the technology behind chatbots like Claude or ChatGPT, requires feeding the system enormous quantities of text or audio. Much of that material is protected by copyright. In most jurisdictions, using it without a licence is illegal unless a specific exception applies.

The United States has a broad "fair use" doctrine that AI companies have relied on heavily in court. Australia has no equivalent. That gap is what Anthropic called impossible to work around.

Australia has been weighing whether to create a new exception. As AI2Day reported on 2 September, the Trump administration backed OpenAI's fair-use argument in a US copyright case, a sign of how much rides on where this legal line gets drawn. Here in Australia, the country has been considering an opt-out model that would let companies train on publicly posted content by default, unless creators actively refuse. Creative industries have pushed back hard, with representatives at this week's hearing arguing any weakening of protections would reduce artists to collateral damage in a race to build AI.

What does this mean for artists and creators?

For musicians and writers whose work lives online, the fight is direct. A broad training exception would mean their published work could be used to build commercial AI products without payment or permission. The opt-out model places the burden on creators to refuse, rather than on companies to ask.

Suno, the AI music company, moved differently: AI2Day covered how it launched its v6 models trained on licensed songs in September, suggesting at least one company sees licensed training as workable. That's the contrast worth watching.

What happens next?

The parliamentary inquiry is ongoing. No vote has been scheduled, and Australia hasn't committed to any specific change.

OpenAI's apology over the Medicare breach, delivered for the second time at the same hearings, adds a separate pressure. The company acknowledged it has "more work to do to rebuild trust with the Australian people." When a government is already scrutinising your data security, arguing simultaneously for looser content rules is a harder sell. It's worth noting, too, that this same committee has been grappling with AI conduct questions since at least August, when AI2Day reported on fake AI-generated research flooding the inquiry's own submissions.

The real question isn't legal. It's whether "we can't build here otherwise" is a reason to rewrite the rules or a reason to build somewhere else.

Common questions

Does this affect Australians who use AI tools today?

Not immediately. The inquiry is examining future rules; existing AI products remain available. What changes at the inquiry could affect whether, and on what terms, AI companies invest in Australian infrastructure.

What is the difference between a copyright exception and an opt-out system?

A copyright exception means a company can use protected work without asking. An opt-out system is a type of exception where creators must actively register their refusal; silence is treated as permission.