Welcome to Neural Notes, a weekly column looking at how AI is impacting Australia. In this episode: Why Australian businesses should pay attention to the German copyright judgment against OpenAI.
Earlier this month, the Munich District Court ruled that OpenAI violated German copyright law by training ChatGPT on nine famous German songs without permission.
The lawsuit was brought by German rights group GEMA, which represents about 100,000 songwriters and publishers. As a result, OpenAI was awarded damages for using the artist’s work to “teach” its model.
The court rejected OpenAI’s argument that OpenAI does not “store” or “copy” individual songs and that users, not the company, are responsible for infringing output. In the judge’s view, the model itself was built on unauthorized material, and that alone is enough to violate the law.
OpenAI disagrees with the ruling and said it is considering an appeal.
GEMA CEO Tobias Holzmuller called it a “landmark decision” that reaffirms that “the Internet is not a self-service store.”
The German Association of Journalists and other rights groups have already hailed this as a turning point, saying it could set the first clear precedent for AI developers to be paid for what they use.
U.S. courts have previously dismissed several author-initiated lawsuits against OpenAI and Meta on procedural grounds, but no court has yet issued a final ruling on whether training on copyrighted data is legal.
Europe has it now. And Australia has also chosen an ally.
After months of pressure from the Technology Council of Australia and other industry groups to adopt the U.S.-style “fair use” exception, the Albanian government waived the exception in late October.
There is no free pass for tech companies to train AI on copyrighted content without license or consent. Instead, the Department of the Attorney General has signaled a move towards a licensing framework and stronger transparency obligations for anyone developing or deploying generative AI systems.
That change has already begun. The ministry’s consultation on AI and copyright, which began in October, is expected to release a draft framework in the coming months, but no formal timeline has been set yet. This may require AI developers to disclose data sources and negotiate with rights holders.
Related Articles Block Placeholder
Article ID: 269876
Australia is effectively aligned with the European licensing approach, rather than the US-style ‘fair use’ obligation, which requires no permission.
Behind the scenes, OpenAI has begun signing multi-year licensing deals with major publishers, and Google and major record labels are negotiating AI-focused music licenses.
Some would argue that this signals that the days of free training data are coming to an end. What happens in Germany will shape the expectations not only of these deals but also of Australian regulators who are watching closely.
Why is copyright in AI an issue?
For companies, this is not a niche legal topic. It’s not just a story of government versus big tech.
Let’s take another look at OpenAI’s argument in Munich that users, not OpenAI itself, are responsible for copyright infringements that occur on its platform.
Although most companies have not built a foundational model, many are already using one. This is true whether you do this through AI capabilities built into productivity software and marketing tools, or directly using something like ChatGPT or Perplexity.
Although the German judgment has made it clear that responsibility is placed up the supply chain, reputational and contractual risks still flow down the supply chain.
If a company relies on AI output derived from unauthorized or stolen material, it will be exposed. And the big tech companies that provide the tools and sell the idea of ”become more productive at all costs” are unlikely to be able to help you when it comes time to actually pay for it.
