The Productivity Commission’s (PC’s) interim report, “Harnessing data and digital technology,” released on 5 August 2025, proposes amending the Copyright Act to Include a Fair Dealing Exception for Text and Data Mining (“TDM Exception”).

The proposal has triggered a firestorm of criticism from rights holders, including musicians, visual artists, and publishers, who argue that the innovation benefits of better informed AI technology do not justify the transfer of property rights. The vociferous reaction makes sense because the AI companies’ conduct has already been questionable; while lobbying for relief from copyright, AI companies are knowingly using copyright material for training without regard for the legalities. 

This report examines whether a new ‘fair dealing’ exemption should be granted to technology companies to train AI models using copyrighted material in order to promote AI innovation. In our view, the Australian Government (and also the New Zealand Government) should be wary of making concessions to the global AI operators. 

Instead of proposing exemptions, the Productivity Commission should promote streamlined marketplaces where buyers and sellers of intellectual property can negotiate mutually beneficial deals – not one-sided exemptions that will primarily benefit foreign technology firms at the expense of local creative industries. 

Executive Summary

• The Productivity Commission’s (PC) interim report, “Harnessing data and digital technology,” published in August 2025, proposes significant changes to Australia’s copyright laws concerning Artificial Intelligence (AI) training. The core challenge identified is that AI models require vast quantities of data, including copyrighted material, for training, and current law typically requires permission from copyright holders.
• A central proposal is to amend the Copyright Act to include a fair dealing exception for Text and Data Mining (TDM Exception), which would apply to all analytical techniques using machine-read material to identify patterns, not just AI training. The PC suggests this would not be a “blank cheque” and use would still need to be “fair”. This exception could particularly benefit smaller, domestic AI models.
• The proposal has been met with a “firestorm of criticism” from rights holders, including musicians, visual artists, and publishers, who view it as a “blueprint for theft” of Australian creative content.
  • Organisations like the Copyright Agency, ARIA/PPCA, MEAA, ASA, and News Corp argue that the proposed changes are unnecessary, contrary to Australia’s interests, and would decimate the value of Australian creative industries by transferring resources to largely foreign big tech companies without compensation.
  • They express dismay that the report acknowledges unconsented use of content by AI companies yet still explores an exemption, arguing it would undermine existing and potential licensing markets.
• Arguments for a TDM exemption highlight its potential to boost Australia’s productivity and economic growth (estimated $116 billion to GDP), drive innovation and foreign investment, facilitate AI model training by simplifying access, support domestic AI development, and align Australia with international trends where similar exemptions exist (e.g., EU, US, Japan, Singapore). Some proponents argue that using content for AI training is “non-expressive” and thus should not constitute infringement.
• Arguments against an exemption also cite concerns that it would legitimise the “wholesale theft” of content, decimate creative industries by eroding creator livelihoods, and undermine licensing markets. There are significant concerns about an “opt-out” model, viewing it as fundamentally at odds with property rights and presenting numerous practical difficulties for creators (e.g., technological nuances, platform control, downstream use, conflict with terms of service). Furthermore, opening free access could undermine incentives for creation, leading to low-quality, AI-generated content.
• An overarching problem identified is the bad conduct of the tech industry in its use of copyrighted digital content, with concerns that a TDM exemption would rapidly become a “blank cheque” due to ineffective enforcement.
• As an alternative, the report advocates for developing and leveraging streamlined markets for copyright material based on licensing, rather than expropriation. This approach:
  • Ensures fair remuneration for creators and provides legal certainty for AI developers.
  • Can be facilitated by collecting societies (like the Copyright Agency), which represent multiple copyright holders and can issue licenses efficiently.
  • Requires transparency from AI developers, including maintaining records of ingested copyrighted works.
  • Allows for safeguards against infringing AI outputs to be stipulated within licensing agreements.
  • Is supported by precedents where major content owners have already negotiated licensing deals with AI companies.

The Productivity Commission’s Interim Report

The Productivity Commission’s (PC) interim report, “Harnessing data and digital technology,” published in August 2025, investigated whether Australia’s copyright laws should be amended to facilitate the training of Artificial Intelligence (AI) models. 

The core challenge identified is that developing and refining AI models requires vast quantities of data, including copyrighted material like web pages, books, images, and music. Currently, using such material for AI training typically requires permission from the copyright holder, as AI models must “copy” this material during their training process. The PC notes concerns that Australian copyright law might not be adequately supporting AI development or that developers are circumventing existing licensing mechanisms.

The PC advocates for a proportionate, risk-based, outcomes-based, and technology-neutral approach to AI regulation, asserting that AI-specific regulations should only be a last resort if existing frameworks prove insufficient or if technology-neutral regulations are not feasible. Instead, the Commission believes that existing copyright law frameworks can be adapted to address the issues posed by AI.

The report outlines several policy options for public feedback, aiming to update the copyright regime:

• No Policy Change: Copyright owners would continue to enforce their rights under the existing legal framework, primarily through the court system.
• Policy Measures to Better Facilitate Licensing: This option explores mechanisms to streamline the process of obtaining licenses for copyrighted materials, such as through collecting societies that represent multiple copyright holders. The Copyright Agency, for instance, has indicated its capacity to assist sectors in using third-party content for AI via annual licenses.
• Amending the Copyright Act to Include a Fair Dealing Exception for Text and Data Mining (TDM Exception): This is a central proposal.
  • Scope: Such an exception would apply to all forms of analytical techniques that use machine-read material to identify patterns, trends, and information, not exclusively AI model training. This would also benefit research sectors for statistical analysis.
  • “Fairness” Requirement: The PC stresses that a TDM exception would not be a “blank cheque”; the use would still need to be “fair” in the specific circumstances to protect copyright holders’ legal and commercial interests. Legislative criteria or regulatory guidance might be necessary to clarify what constitutes fair use.
  • Implementation Options: The PC is seeking input on whether the exception should be a broad TDM exception or limited to non-commercial uses only. An “opt-out” for copyright holders is another option.
  • Potential Effects: The PC suggests a TDM exception is unlikely to impact the availability of large AI models in Australia (as many are trained overseas) but could significantly benefit smaller, low-compute models developed domestically by Australian research institutions and medical technology firms.

The Commission notes that several countries already provide similar exemptions for text and data mining. However, many of these are narrowly focussed:

• European Union: The EU has incorporated two Text and Data Mining (TDM) exceptions within its Digital Single Market Directive (EU 2019/790): one for scientific research (Article 3) and another for general use (Article 4). The AI Act (Regulation (EU) 2024/1689) explicitly characterises AI model training as involving “text and data mining techniques” and refers to these existing TDM exceptions. The EU Parliament has supported mandatory TDM exceptions for research and broad optional exceptions for information analysis. The AI Act also includes transparency and copyright-related obligations for General-Purpose AI (GPAI) model providers, requiring them to adhere to rightsholders’ “opt-outs” and publicly summarise training content, even if the training occurs outside the EU, to ensure a level playing field.
• United States: The “fair use” doctrine allows for incidental copies of works for informational analysis without infringement, even for commercial purposes. Whether AI training falls under fair use depends on the specific circumstances and the application of four factors. The U.S. Copyright Office has noted that various uses in AI training are likely transformative, but their “fairness” is context-dependent.
• United Kingdom: There is an existing TDM exception for non-commercial research. Proposals to expand this exception to cover all uses are currently under consideration. The UK Government is assessing the economic impact of four policy options, including broadening the TDM exception with or without an opt-out mechanism.
• Japan: Japan’s Copyright Act includes broad statutory exemptions for TDM (Article 30-4(ii)), provided the work is used for “non-enjoyment” purposes, essentially distinguishing between consumption as a work versus as data.
• Singapore: Singapore also has a specific TDM exception, in addition to a broader fair use exception.

Cultural industries react

The Productivity Commission’s interim report has faced significant opposition from various Australian creative organisations, publishers, and cultural industry stakeholders, many of whom have rejected its proposals outright. The overarching sentiment is that the proposed changes amount to a “blueprint for theft” of Australian creative content.

• The Copyright Agency, a not-for-profit body responsible for distributing royalties, declared that “it is not necessary, or in the interest of Australians, to change Australia’s copyright regime to benefit multinational tech companies”. They highlighted their capacity to facilitate licensing for AI-related activities, suggesting existing frameworks could be leveraged.
• The Australian Recording Industry Association (ARIA) and the Phonographic Performance Company of Australia (PPCA), which safeguard the interests of musicians and songwriters, called the proposal “ill-considered and contrary to Australia’s best interests”. Their CEO, Annabelle Herd, warned that “granting technology companies unrestricted access to exploit generations of Australian artistic and cultural output will decimate the value of Australian creative industries and place our creators at a disadvantage internationally.” She urged the PC to “work to optimise existing licensing frameworks that can deliver promised AI productivity gains without gutting Australian copyright”.
• The Media, Entertainment and Arts Alliance (MEAA) also issued a strong rejection, describing the report as “a blueprint for the wholesale theft of Australia’s art, media, and cultural heritage that will do nothing more than further enrich the billionaires in Silicon Valley”. MEAA chief executive Erin Madeley stated that any rollback of copyright law would “effectively enable a transfer of resources from Australian creatives to mostly foreign big tech companies, with no chance of compensation.” 
• The Australian Society of Authors (ASA) expressed deep dismay, noting the report acknowledged that Australian creators’ work had already been used for AI training without consent or compensation, yet still explored an exemption. ASA CEO Lucy Hayward stated, “Copyright is how authors earn a living. A text and data mining exception would give tech companies a free pass to use their work to train artificial intelligence models – and profit from it – while Australian creators get nothing. Not only is that absurd, it’s unjust.” She found the “not a ‘blank cheque’ ” argument “unconvinced” and argued that TDM exceptions “don’t benefit creators, Australian culture, or even tech companies, who need writers and artists to survive and continue to create high-quality books to develop and improve their AI tools.” Instead, she contended, it “undermines existing and potential licensing markets.”
• News Corp, through its executive chairman Michael Miller, strongly criticised the report, stating, “The harm is real with Australia being asked to trade away our cultural, social and economic sovereignty despite no genuine evidence that Australia’s copyright laws are stifling innovation or investment.” Miller challenged large technology companies, asking, “If big tech wants free and open access to other people’s intellectual property are they prepared to give us free and open access to theirs?”.

From a tech industry perspective, Scott Farquhar, co-founder of Atlassian, publicly advocated for an “urgent” overhaul of Australian copyright laws, believing they were stricter than those in the UK and US, thus hindering Australia’s ability to compete in AI development. He suggested that creating exemptions for text and data mining to train large language models “could unlock billions of dollars of foreign investment into Australia”. 

In contrast, the Assistant Minister for Productivity, Andrew Leigh, has indicated support for the PC’s general approach of adapting existing regulations rather than creating overarching AI-specific legislation. 

PC Commissioner Stephen King acknowledged the “obvious harm” of uncompensated use of copyrighted material by AI companies but also the need to develop AI tools that use such material, suggesting existing copyright collection models might be relevant.

Is a TDM justified?

So far, the debate surrounding a copyright exemption for AI training is fundamentally an economic one, weighing the potential for innovation and productivity gains against the protection of existing creative industries and the livelihoods of creators. 

But these pros and cons are subsumed by a larger issue – the history of bad conduct of the tech industry. There is no serious dispute that AI training has already led to significant use of copyright material without compensation, and this is also true in other contexts like search and social media. A TDM assumes that the tech industry would respect the boundaries set by policy – but will they?

Arguments for an Exemption 

The PC views AI as a “new wave of productivity growth” and a solution to “anaemic productivity growth”. Early estimates suggest AI could increase Australia’s multifactor productivity by at least 2.3% over the next decade, translating to an estimated $116 billion boost to GDP or a $4,300 increase in real wages per capita. An exemption is seen as crucial for unlocking these benefits:

• Driving Innovation and Investment: Proponents like Atlassian co-founder Scott Farquhar argue that current Australian copyright laws are too restrictive, preventing Australian companies from competing globally in AI development. He believes that granting TDM exemptions “could unlock billions of dollars of foreign investment into Australia”. A clear exemption would reduce “regulatory uncertainty,” which can “stifle innovation and investment” if firms fear onerous or unclear regulations.
• Facilitating AI Model Training: AI models require “massive amounts of data” for training. An exemption would simplify the process of legally accessing and copying copyrighted material, which is currently a barrier due to the need for permission.
• Supporting Domestic AI Development: While large AI models are often trained overseas, a TDM exception could be particularly beneficial for smaller, low-compute models built and trained domestically by Australian research institutions and medical technology firms, fostering local innovation.
• International Alignment: Many comparable jurisdictions, including parts of the EU, the US (through fair use), Japan, and Singapore, have TDM exceptions or similar doctrines. Failing to align with these international trends could make Australia less attractive for AI talent and investment, causing it to “lag in AI take-up” and potentially undermine its AI sector.
• Non-Expressive Use Principle: A core argument for TDM exceptions is that the use of copyrighted material for AI training is “non-expressive.” Copyright typically protects the expression of ideas, not the underlying information or data itself. From this perspective, using content to identify patterns for AI training should not constitute infringement.

Arguments Against an Exemption 

Creative organisations vehemently reject the proposed TDM Exemption, viewing it as a legalisation of “wholesale theft”. They argue it would allow “the wealthiest corporations in Big Tech” to “freeload from low-paid authors’ labour,” effectively “institutionalising wage theft”. They emphasise that AI models are already being trained on copyrighted material without consent or compensation, and an exemption would simply legitimise this. Key issues include:

• Erosion of Creative Industries and Livelihoods: Copyright is fundamental to how creators earn a living. An exemption would give tech companies “unrestricted access to exploit generations of Australian artistic and cultural output,” which, according to Annabelle Herd of ARIA/PPCA, “will decimate the value of Australian creative industries and place our creators at a disadvantage internationally”. This could lead to a “transfer of resources from Australian creatives to mostly foreign big tech companies, with no chance of compensation”.
• Undermining Existing and Future Licensing Markets: Opponents argue that a TDM exception would “undermine existing and potential licensing markets” that currently provide a means for tech companies to access content while remunerating creators. They believe copyright law, when properly applied, should set conditions for the market to prevail, allowing creators to choose whether to license their works for AI purposes.
• Erosion of Property Rights (Opt-Out Model Concerns): The idea of an “opt-out” model, where creators must explicitly forbid their work from being used, is seen as “fundamentally at odds with the notion of property”. Copyright, like other property rights, should be inherent and not require active steps to retain. This is particularly problematic given that generative AI is already beginning to replace human works.
• Practical Difficulties of Opt-Out: Implementing an effective opt-out system presents numerous challenges:
  • Technological Nuances: Watermarks or hashtags like “#NoAITraining” might not be legally effective or machine-readable. European examples suggest “machine-readable means” (e.g., specific metadata, robots.txt files) are necessary, which can be technologically complex for individual creators.
  • Platform Control: Web-crawler blockers are bot-specific, page-specific, and not work-specific, meaning they can be bypassed or require constant updates. Metadata can be easily altered, removed, or stripped by platforms like social media sites.
  • Downstream Use: Opt-out notices might be ineffective for “downstream” re-publications or derivative works, meaning creators could lose rights when their work is shared beyond their control.
  • Existing Works: The model raises questions about whether existing online content without prior opt-out notices would become “fair game” for AI training.
  • Conflict with Terms of Service: Platforms (many of which are developing their own AI) might have terms of service that conflict with opt-out requests, potentially penalising users who opt out.

Even more fundamentally, there is a risk that opening up free access to content resources will undermine long-term incentives to create such content. This could lead to a race to the bottom where low quality, AI-generated works proliferate, further crowding out human creation. AI models continuously trained on AI-generated content can produce low-quality, incorrect, or biased outputs. As incentives for creation are eroded, human creative intensity is diminished, and the entire ecosystem is undermined.

In summary, an exemption could boost Australia’s AI sector by providing cheaper, easier access to training data and attracting foreign investment. However, it risks disadvantaging a significant portion of Australia’s creative and cultural industries, who see it as a fundamental erosion of their intellectual property rights and a threat to their livelihoods. Finally, it would erode the incentives for creation, ultimately undermining the information ecosystem on which AI depends.

The overarching problem – oligopoly power and bad conduct

AI technologies will undoubtedly create a great deal of economic value over the next decades, but this is not the only consideration when deciding their access to copyright material. There is also the question of who benefits. The immediate benefits accrue to AI companies themselves, who gain free access to training material. Others benefit from the innovation and functionality that AI technologies then bring.

In an efficient, competitive market others would benefit significantly, reaping a large consumer surplus. The problem however, is that the market for AI technology more resembles an oligopoly where AI companies exercise significant market power as their scale grows. These oligopolies are loss-leading their AI investments to lock the majority of customers into their technology ecosystems, by tightly integrating their existing software services with AI functionality, and in some cases vertically integrating into the datacentre compute power needed for AI (Google and Microsoft stand out).

Their current focus is market share, but this oligopoly power will give them the opportunity to raise real prices over time. We have seen just such a trend in cloud services, where Google, Microsoft Azure, and Amazon AWS dominate. This would allow the AI tech companies to capture an outsize share of the total value creation, reducing benefits to nations like Australia that lack a large AI industry.

And there is also the problem of tech company culture. Given the tech industry’s reputation for abuse of copyrighted digital content, why should anyone believe that the industry would keep access within any bounds legislators may set? It is much more likely that a TDM would rapidly become a blank cheque – despite PC claims to the contrary. And since there have been few consequences for the past breaches, tech platforms would not expect any consequences for breaching the limitations that the PC envisages. The platforms’ well-documented reluctance to cooperate with copyright holders over content re-use in the context of search and social media is an indicator of this larger cultural problem.

The reality of the tech industry’s bad faith makes the construction of any effective TDM very difficult. It would need to be highly prescriptive to avoid manipulation. And even if legislators successfully closed all of the loopholes, experience shows that tech companies are prepared to harvest publicly available data indiscriminately, knowing that non-compliance is hard to prove. Further, the burden would mainly be on copyright holders and their representatives to demonstrate such breaches, making it even harder to enforce. 

The result would be a significant loss of value to copyright holders, a significant gain to tech companies, but uncertain net gains for anyone else. At the very least, proponents of a TDM need to explain how the problems of oligopoly power and bad faith can be addressed.

Streamlined markets as an alternative

Instead of a broad copyright exemption for AI training, we argue for developing and leveraging more streamlined markets for copyright material. This approach relies on licensing, rather than expropriation, as the mechanism for AI developers to gain access to the vast amounts of structured data they need, while simultaneously ensuring creators are fairly remunerated and retain control over their intellectual property.

The problems of oligopoly power and bad faith remain. However, a licensing approach mitigates this in two ways. First, access is still available – but not for free, providing an alternative to exploitation. Second, there is no exemption that can be “creatively” abused to allow free access. While this would not prevent the outright theft of content, it does not obscure the fact that it is indeed theft. That brings added risk to tech companies that encourages a lawful approach.

The role and benefits of licensing

• Primary Mechanism for Permission: Licensing is highlighted as the “key mechanism” through which copyright holders grant permission for others to use their work, typically involving some form of payment. This ensures that creators are compensated for the value derived from their intellectual property.
• Fair Remuneration: Organisations representing creators, such as the ARIA and PPCA, argue that existing legal frameworks already provide clarity for “licensing negotiations and balanced agreements that fairly reward creators and give them control over exploitation of their works”. They advocate for optimising these existing frameworks rather than introducing exemptions that would “gut Australian copyright”. The Copyright Alliance explicitly states that “AI companies should license works they ingest” because copyrighted works “provide immense value to AI developers, and they can and should pay for that value”.
• Avoiding Infringement Liability: Obtaining a license is considered the “best way for developers to ensure they avoid infringement liability”. This provides legal certainty for AI developers, potentially reducing future litigation costs and risks.
• Market-Based Solutions: If licensing markets are already established or are being developed, it can mitigate the need for “fair use” or “fair dealing” exemptions. The existence of a functioning market for content suggests that creators are willing to license their work for appropriate compensation, negating the argument for free access due to market failure.
• Precedent for Negotiation: Some major content owners, like NewsCorp, Shutterstock, and Conde Nast, have already chosen to negotiate and execute licensing deals for their content with AI companies, demonstrating that market-based solutions are viable.

Mechanisms for streamlining access via licensing

The Collecting Societies can play a crucial role in streamlining the licensing process. By representing multiple copyright holders, collecting societies can negotiate and issue licenses on their behalf, making it easier and more efficient for AI developers to obtain the necessary permissions without having to approach individual creators. The Copyright Agency in Australia exemplifies this, stating they can assist with licensing third-party content for AI and are exploring “collective licensing solutions”. Other potential benefits include:

• Transparency and Record-Keeping: To facilitate effective licensing and address concerns, the Copyright Alliance emphasises the importance of transparency from AI developers. This includes “maintaining records of what copyrighted works are being ingested and how those works are being used,” and making these records “publicly accessible and searchable as appropriate”. Such transparency helps copyright owners understand if their work has been used and can inform licensing negotiations.
• Integration of Safeguards in Licensing Agreements: Licensing agreements can be used to stipulate safeguards that address concerns about the outputs of AI models. For instance, negotiations could include provisions to prevent infringing AI-generated outputs or to manage issues like “overfitting” or “in the style of” prompts, ensuring responsible AI development.
• A “Fair Exchange” Principle: The Australian Society of Authors (ASA) advocates for a “fair market” which necessitates a “fair exchange”. This implies that if AI companies derive value from copyrighted works, that value should be shared back with the creators through remuneration.

As we recently noted (see our report “The End of Search: Navigating the Transition to an AI-Centric Information Ecosystem”), this approach of selling structured data sets on an open market for data will characterise the AI ecosystem in the future. 

This will require the collecting societies to develop these structured data sets and the associated marketplaces – a significant challenge. However, this is also a task for policy-makers, who will need to create regulatory frameworks to support the emergence of these markets.

In summary, we argue that focusing on and strengthening licensing frameworks, underpinned by transparency and the collective bargaining power of collecting societies, presents a more equitable and sustainable path forward for AI development in Australia than a TDM exemption which would cripple intellectual property markets and the associated creative industries. This approach aims to foster innovation while ensuring the continued vitality and compensation of the creative industries.

About Venture Insights

Venture Insights is an independent company providing research services to companies across the media, telco and tech sectors in Australia, New Zealand, and Europe.

For more information go to ventureinsights.com.au or contact us at contact@ventureinsights.com.au.