Copyright, AI, and the Public Domain:
A Roadmap for Indonesian Storytellers

Australia occupies an interesting position: it is proactively engaged with questions of AI and public domain, but has not yet reached any legislative resolution. Its copyright framework follows the standard international model life plus 70 years for works with a known author, publication year plus 70 years for anonymous works, and 50 years from broadcast year for broadcasts. A notable quirk: government works automatically enter the public domain unless otherwise stated, creating a meaningful resource base for creators and researchers.

This became particularly significant following the 2005 Australia–US Free Trade Agreement, which extended copyright duration from life plus 50 to life plus 70 years aligning Australia with the US and EU, but also delaying the entry of many works into the public domain. The practical consequence is illustrated by a familiar example: Banjo Paterson, author of "Waltzing Matilda," died in 1941, meaning his work entered the public domain in 1991 under the old rules. Had the extended term applied, it would have been 2011 instead.

On AI, Australia's position is nuanced and still evolving. The Copyright Agency summarises it thus: Australia's copyright legislation applies to AI-related activity as a matter of general application there are currently no AI-specific provisions. Whether copyright can be claimed on AI-generated works is considered on a case-by-case basis, contingent on whether sufficient creative effort was invested by a human author. This approach prioritises flexibility over certainty, which is useful for innovators but creates genuine legal ambiguity for platforms like Librery.

If you want to understand where the regulation of AI and public domain is heading globally, you look to Europe. The EU has spent decades harmonising copyright law across its member states, and is now applying that same methodical rigour to the governance of artificial intelligence. The result is a framework that is simultaneously the most complex, the most principled, and for builders like the Librery team the most instructive.

The standard copyright duration in the EU is life plus 70 years, applied uniformly across member states following decades of harmonisation. For audiovisual works, the 70-year term runs from the death of the last surviving key contributor whether principal director, screenplay author, dialogue author, or original music composer. Performers and producers receive 50 years of protection. Government document policies vary by country, though the European Commission has consistently pushed for open access to publicly funded research.

Europeana stands as the EU's flagship public domain repository: a vast digital aggregation of materials from European museums, libraries, and archives, explicitly committed to public domain preservation through the Europeana Public Domain Charter. This is complemented by a pilot centralised repository for openly licensed works, and an extensive open access mandate for Horizon Europe-funded research.

On AI, Europe's landmark contribution is the EU AI Act the world's first comprehensive legal framework for artificial intelligence. Rather than a narrow copyright statute, it establishes a risk-tiered approach to AI governance that explicitly engages with questions of creative rights, data provenance, and public domain use. The EU's approach treats cultural heritage as a public good, not merely intellectual property a philosophical orientation that has direct implications for how public domain materials can be used in AI training datasets.

The United States presents the most dramatic case study in how copyright law can be shaped by commercial interests rather than creative ones. The pivotal moment came in 1998, when the Sonny Bono Copyright Term Extension Act widely referred to as the "Mickey Mouse Protection Act" extended copyright duration by 20 years, just as Steamboat Willie (1928) was approaching the public domain. Disney's lobbying was instrumental; the extension delayed Steamboat Willie's entry into the public domain until January 2024, a full 95 years after its creation.

The irony is instructive. Mickey Mouse's first appearance is now in the public domain but only in its 1928 form, without pants. Derivative versions, sound recordings, and the modern character remain protected. Independent developers and game creators have immediately begun using the original Steamboat Willie, generating a wave of creative (and sometimes surreal) derivative works that demonstrate exactly what public domain access enables.

For works created by individuals, the US standard is life plus 70 years. For works-for-hire, anonymous, and pseudonymous works typically corporate productions the duration is 95 years from publication or 120 years from creation, whichever is shorter. Federal government works, by contrast, are immediately in the public domain: NASA images, Supreme Court rulings, and congressional records are all freely available from the moment of creation.

The US approach to AI and copyright is still crystallising. The Copyright Office ruled in 2023 against granting copyright to purely AI-generated works (in the case of AI artist Kris Kashtanova). Getty Images has sued AI companies for using copyrighted photographs in training datasets without permission. These cases are establishing precedent in real time and their outcomes will reverberate globally, including in Indonesia.

Japan is perhaps the most quietly interesting jurisdiction in this survey. While it aligns with international standards on copyright duration extending from life plus 50 to life plus 70 years in 2018, following the Trans-Pacific Partnership it has taken a notably pragmatic approach to AI and public domain use that makes it, in some respects, the most permissive major economy for AI training.

Under Japanese copyright law, AI can be trained on copyrighted material if it meets certain conditions specifically, that the training does not amount to imitation or similarity of the original work, and that the use is not a direct reproduction for consumption. This is a meaningful distinction: Japan's courts have recognised a difference between using a work for informational extraction (as in machine learning) versus reproducing it for a human audience. The caveat is that copyright assessments are made case by case, weighing two factors: whether the person using the AI had a "creative intention," and whether they made a "creative contribution" to the resulting work.

Japan's institutional infrastructure is formidable. The National Diet Library Japan's national library holds over 47.5 million catalogued items, of which 5.5 million have been digitised and nearly 2 million are accessible via the internet. Governmental and institutional publications, including court judgments and rulings, are not protected by copyright. And Japan actively encourages international IP development on its soil: works by foreign creators receive protection if published in Japan within 30 days of international release.

The challenge, as the team notes, is that Japan's framework for AI-generated content remains genuinely unclear particularly around the ownership of outputs and the boundaries of imitation. What the law says is largely what it has always said about human creativity; it simply has not yet been rewritten for the AI era.

India's approach to the public domain is both inspiring and instructive particularly for Indonesia, which shares many of the same structural challenges. Just as India has led the world in digitising financial payments infrastructure, it has poured remarkable ambition into digitising its cultural and academic heritage. The scale is staggering: the National Digital Library of India (NDLI) hosts over 115 million pieces of content across 452+ languages, supports more than 150 institutional repositories, and anchors a network of over 6,400 learning clubs. The Public Library of India, maintained by Public Resource, provides open access to more than 8 million books, audio files, and government publications.

India's copyright framework is deliberately shorter than the Western standard: life plus 60 years, reflecting a conscious decision not to extend terms to life plus 70 as Europe and the US have. This means Indian works enter the public domain earlier a policy choice that prioritises access and education over extended commercial protection. Notably, this creates cross-jurisdictional complexity: the works of Rabindranath Tagore, who died in 1941, entered the public domain in India in 2001 but remain copyrighted in countries with life-plus-70-year terms until 2011.

The most consequential recent development is the ANI vs. OpenAI case one of the first instances of an Indian court being asked whether training AI on publicly available content constitutes copyright infringement. India's courts have previously held that a derivative work can attract new copyright protection if it contains sufficient originality: the source material is reproduced and transformed into something meaningfully different. Whether that logic extends to AI training is the open question the ANI case will help answer.

India's critical lesson for Indonesia: it is possible to build extraordinary public domain infrastructure at scale, even without having resolved every legal question around AI. The NDLI, Shodhganga, and INFLIBNET exist and function and they do so under Creative Commons licensing frameworks (particularly CC BY-NC 4.0) that explicitly enable non-commercial AI use. Indonesia can follow the same infrastructure-first, regulation-evolving model.

Indonesia presents the most complex and, for Poddium's purposes, the most consequential picture. It is a country of extraordinary cultural depth home to hundreds of distinct literary traditions, oral histories, musical forms, and artistic practices. Yet paradoxically, some of the largest collections of Indonesian heritage materials are not in Indonesia at all. The National Library of Australia, through its Indonesian Acquisition Office in Jakarta, holds over 200,000 monographs, 5,000 serials, and 250 newspaper titles. Ohio University in the United States houses the Indonesian Music Archive, containing 193 hours of historical audio recordings. Within Indonesia, ARSIP (Arsip Nasional Republik Indonesia) has spearheaded major digitisation efforts but many collections remain fragmented, inaccessible, or physically housed overseas.

Indonesia's copyright framework is, on paper, one of the more sophisticated in Southeast Asia. Law No. 28 of 2014 the current governing statute established protection for the life of the author plus 70 years for literary, musical, fine art, and spoken works. Joint works extend this to 70 years after the death of the last surviving author. Works owned by legal entities, as well as photographs and cinematographic works, are protected for 50 years from first publication. Applied art works receive 25 years.

Importantly, the 2014 law introduced a set of provisions that are directly relevant to AI-assisted storytelling. It established fair use exceptions for education, research, and limited non-commercial digital use. It recognised moral rights for the first time persisting after death and enforceable by heirs. And it introduced a 25+2 year reversion clause for flat-sold rights, mandating royalty-based licensing going forward. On the institutional side, Satu Data Indonesia a 2019 regulation mandates that all government data be standardised and publicly accessible, creating the foundation for a more robust open data ecosystem.

Yet the gaps are significant. Indonesia has no legal definition of "public domain" and no official registry. Even when works are technically free to use, they are often not digitised, not accessible online, or too difficult to locate. Digital reuse of public domain content is poorly regulated, and public understanding of open licensing is low. A growing risk of "re-privatisation" where institutions digitise public domain works and then claim copyright over the digital versions further threatens open access.

On AI, the situation is both unsurprising and telling. Indonesia currently has no dedicated AI law. The existing framework relies on three statutes: Law No. 27 of 2022 (personal data protection), Law No. 8 of 2011 (electronic information and transactions), and most significantly, Law No. 9 of 2023 ethical guidelines for AI use. That law explicitly states that AI must align with IP protection principles as defined by Indonesian copyright law. But it does not answer the questions that matter most for a platform like Librery: Can AI train on old Javanese folklore? Can it adapt stories from the Mahabharata? Does a human-AI co-written narrative qualify as an original work?