Key Takeaways

  • Australia’s federal government has ruled out a broad AI text and data mining exemption, reinforcing a licensing-first model.
  • Global AI adoption pressures are rising, with more than 80% of enterprises expected to use generative AI by 2026.
  • The lack of a TDM exception places Australia among the most restrictive jurisdictions on AI training data access.

Australia’s ambition to become a major AI producer is colliding with a legal framework first drafted in 1968. Tens of billions in investment, both domestic and inbound, now hang on how policymakers balance copyright protections with the accelerating need for machine-readable training data. The debate has intensified as the government’s explicit rejection of a broad text-and-data-mining (TDM) exemption establishes a defining line in the sand.

The decision set off immediate reactions because enterprises in Australia are moving faster into generative AI deployment than many anticipated. According to Gartner 2024 estimates, more than 80% of enterprises worldwide will have integrated generative AI APIs or apps by 2026. That is not a surprising figure to many CIOs, yet the policy environment in Australia has not kept pace, creating a unique tension between investment appetite and regulatory caution.

Copyright bodies have been raising alarms for years. The World Intellectual Property Organization reported in 2023 that 70% of surveyed copyright offices saw AI training on protected works as a priority uncertainty. And that report matters because Australia often looks to international norms when adjusting copyright policy. It also helps explain why Australia’s rejection of a TDM exception landed with such force. Until now, many in the AI sector believed a more permissive regime was still possible.

Not everything about the current conversation is new. Australia has long been a Berne Convention member, so cross-border copyright protections are not unfamiliar. What has changed is the scale and speed of modern data ingestion. Generative models require massive corpora to reach commercially useful performance. That includes text, audio, images, video, and code. The government’s position, which leans toward licensed access rather than open scraping, sends a clear message to model developers such as OpenAI, Google, and Anthropic.

Another dimension of the issue is the global policy race. The OECD’s 2024 AI Policy Observatory found that more than 30 countries have updated national AI strategies with explicit positions on copyright and data access, placing Australia among the more restrictive regimes on TDM. That framing matters for investors deciding where to place large-scale AI training infrastructure, especially as nations in Asia and Europe shift toward more nuanced or conditional data mining provisions.

Some might ask whether the lack of a TDM exception will slow domestic innovation. It could, although the reality is more complicated. Licensing-based systems offer clearer economic participation for rights holders, which can strengthen the creative sector. On the other hand, licensing at scale can be expensive and difficult to negotiate. Getty Images has pursued litigation around AI use of copyrighted images in other jurisdictions, spotlighting how contentious these relationships can become if licensing arrangements fail.

A different but related pressure point is enterprise governance. The NIST AI Risk Management Framework 2023, widely referenced in industry, encourages organisations to document data provenance and potential copyright exposure as part of their risk assessment controls. In practice, Australian companies will need far more visibility into the training data behind commercial models or choose to build their own curated datasets. Both pathways require investment and new workflows.

For economic planners, the stakes are high. McKinsey’s 2023 estimate that generative AI could add up to US$4.4 trillion annually to global GDP continues to circulate in policy conversations. Australia wants part of that upside. Yet without clearer legal footing on training data, model developers and enterprise buyers may adopt more conservative approaches, particularly in heavily regulated fields such as health, finance, and government services.

The policy environment is not purely defensive, though. A licensing-first system can encourage commercial partnerships between rights holders and model developers. It also aligns with the broader creative industries, which have historically pushed for stronger protection against unlicensed digital use. Some publishers see Australia's stance as a chance to shape a sustainable revenue model for AI-era content use. Whether that optimism holds will depend on how the government implements actual mechanisms for collective licensing or statutory schemes.

On the global stage, Australia’s position puts it closer to jurisdictions experimenting with stricter copyright interpretations. Others, such as parts of the European Union, have allowed TDM exceptions in certain contexts. Businesses comparing these regimes will be weighing predictability, compliance costs, and operational flexibility. The interesting question is whether Australia’s choice becomes a differentiator or a deterrent. For some organisations, certainty has value even when the rules are tight.

Then there is the practical matter of enforcement. Modern AI models often use international data pipelines, where training work takes place across multiple jurisdictions. The Berne Convention offers shared principles, but cross-border enforcement is rarely simple. Policymakers in Canberra are aware of these dynamics, and the next phase of consultation is likely to focus on how to implement licensing structures that can actually function at machine scale.

The conversation is still evolving. Investors want clarity. Rights holders want compensation. AI developers want access. And enterprises, navigating global competitive pressures, just want reliable guidance so they can plan multi‑year technology strategies without fear of retroactive legal exposure. The tension reflects a broader global moment where copyright law, written in a very different era, is being asked to stretch over a rapidly expanding AI economy. Australia’s decision not to adopt a TDM exception signals that the country is choosing control and compensation. Whether that choice unlocks confidence or slows momentum remains to be seen.