There is an urgent need for transparency regulations for Artificial Intelligence (AI) platforms exploiting creative works to train generative models.
From our landmark AI and Music Survey, over 4,000 APRA AMCOS members clearly articulated the necessary actions to be taken.
We round up what is happening in other territories and how it matters here.
As we head into the Federal Election in Australia, we urgently call for transparency regulations for Artificial Intelligence (AI) platforms that exploit creative works to train generative models.
“AI is already reshaping the way music is created, consumed and monetised, often without the consent or knowledge of the people who make it,” said Dean Ormston, APRA AMCOS CEO. “We need clear, enforceable rules around the training of AI models, including transparency about what content is being used. Without this, Australian and New Zealand creators risk being exploited, their works devalued and their voices erased.”
From our landmark AI and Music Survey, over 4,000 APRA AMCOS members clearly articulated the necessary actions to be taken.
Policy demands that you want: 97% demand that policymakers should pay more attention to the challenges related to AI and copyright; 96% credit and transparency, 95% consent and 93% remuneration.
The AI and Music report has driven greater public awareness of the issue through over 300 news stories in Australia, New Zealand and beyond, including The Australian’s Review cover story (paywall article) featuring Nick Cave.
On the topic of generative AI, Cave said, “That's its intent: to completely sidestep the inconvenience of the artistic struggle (by) going straight to the commodity. Which reflects on us, as what we are as human beings: things that consume stuff. We don't make things anymore; we just consume stuff.”
“93% of copyright holders saying that they should be remunerated for their works is a really amazing response, very powerful in terms of us being able to show to the government,” said Ormston.
“The key point really is to make clear that generative AI platforms, the large language models, multinational platforms, have already ingested the world’s creative content. They've converted that content, created content into data points.
“It's stored and repeatedly accessed in terms of training large language models. And repeatedly, that creative content as data points are accessed for the generation of new AI outputs.
“So, there's now an entire economic ecosystem of businesses that are making money out of Generative AI, and the creative community is yet to make a dollar.
“The response from bureaucrats in Australia and New Zealand in terms of the government opposition and crossbenches has been really positive. People are understanding.
“The Australian government produced a report that has a number of recommendations relating to mandatory guardrails for AI, and we welcome this report.
“It specifies that there should be safeguards for high-risk AI activities, and user-generated content is considered a high-risk activity due to its potential impact on the businesses of creators.
“The government has called out that there needs to be transparency, that platforms need to make clear what data they have ingested.”
Here’s a round-up of the key international developments involving creative industries and generative AI technology.
In December 2024, CISAC published the first-ever global study measuring the economic impact of AI in the music and audiovisual sectors.
Of the key litigation cases around the world, we are closely watching what happens in the US, where most of the cases are taking place.
A significant judgment in the case opposing Thomson Reuters vs. Ross found that the generative AI platform in question made reproductions of copyrighted materials and that the ‘fair use’ defence did not apply.
The EU's AI Act introduces transparency obligations for generative AI platforms, with ongoing debate on the required level of detail of disclosure proposed in the third draft of the EU AI Act’s General Purpose AI (GPAI) Code of Practice.
CISAC recently signed a joint statement initiated by a coalition of authors, performers, and other rightsholders' organisations regarding the GPAI Code of Practice, expressing strong concerns from both the EU and international cultural and creative sectors about the latest draft, which fails to include the necessary measures and commitments to ensure effective copyright compliance and transparency for GPAI models.
Importantly, the proposed Key Performance Indicators (KPIs) and transparency measures, as currently outlined, risk weakening the provisions of the EU AI Act and undermining the primary goal of the Code of Practice, which is to ensure compliance with the AI Act and other EU laws.
There's currently a lot of discussion about proposals to introduce new exceptions into the Copyright Acts in Hong Kong and the United Kingdom called ‘text and data mining exceptions’ and they're more specific than the fair use exception that exists in the US.
These exceptions would allow generative AI platforms to scrape, mine, and copy – essentially ‘feed’–copyrighted works from the internet into their AI models without requiring the platforms to seek a licence or provide remuneration.
“The plan to grant an exemption to UK copyright rules for AI companies looking to train their models was announced earlier this year as part of a cross-government push to boost the British AI industry,” a Guardian article explained.
There has been pushback from creators in the UK to the proposal, with the silent album Is This What We Want? highlighting the detriment to artists livelihoods that could result from the policy, which would put the onus on copyright holders to ‘opt out’ of the ‘text and data mining’.
One of the key lobbying and advocacy points we have been addressing with the Australian and New Zealand governments is that exceptions of this nature should not be introduced in our territories.
Point of difference
A really interesting point of difference between the proposed Hong Kong exception and the UK’s is that in the Hong Kong bill, the text and data mining exception would only apply when no licensing solution exists.
With this in mind, it becomes more important for APRA AMCOS and other rightsholders to explore building the necessary rights mandate to offer a collective voluntary licensing solution in Australia and New Zealand and to be prepared for different policy and business scenarios so as to avoid the introduction of any free use exception in our territories.
APRA AMCOS’ German affiliate GEMA has brought litigation against two music-specific generative AI platforms, Suno AI and OpenAI. The outcome of this litigation could impact generative AI platforms' approach to managing their risk and potential liability for copyright infringement when training their large language models.