Whatever the outcome, this is where the licensing-led approach gets stress-tested in a major emerging market, with the usual suspects lined up against it and the real design questions still open.
A bill due before parliament by 12 December could make SE Asia’s largest market the clearest test yet of a licensing-led answer to generative AI. English-language coverage has been sparse and unreliable (some sources could not differentiate between ASEAN and APAC), so I’ve relied on my AI buddies for translations from Indonesia media sources. With that caveat, here is what is known, what is inferred, and what nobody yet knows.
Such English-language coverage there is of Indonesia’s proposed copyright revision largely follows Reuters’ account of an 11 November parliamentary hearing, and frames it as a fight over whether Google, Meta and TikTok should pay news publishers. An easy headline that buries the real story.
Indonesian-language reporting on the same hearing, and on the weeks of committee work before it, describes something broader (as above, AI translation helped me get there): A a bill that reaches authorship, training data, translation, lending rights, collective licensing and platform liability. For anyone who publishes books, journals, translations or educational content, in Indonesia or into it, that is a very different story from a dispute over news links.
The TNPS analysis tries to keep three things apart. Known means reported by named officials, companies or outlets. Inferred means my own reading of that evidence, so have afield day with that! Unknown means what the reporting does not tell us, and in places that is a great deal. I’ve skipped the unknown unknowns, as by definition I don’t know what they are, and even Claude couldn’t help me there.
What is known
Reported by named sources. I make no pretence of having seen or read the bill, so treat this as “apparently”.
The process. The bill amends Law No. 28 of 2014. A plenary sitting of Indonesia’s parliament set it as a House initiative bill on 12 March 2026, and formal deliberation with the government began on 6 October, when the government submitted a 1,117-point list of issues to resolve. The sponsors want a plenary vote before the session closes on 12 December.
The AI provisions. According to the draft Reuters reviewed in July, works made with AI assistance can be protected if they meet human-involvement criteria, while fully AI-generated works are excluded. The draft bans AI imitation of a creator’s distinctive style, requires disclosure of AI use, and permits training on copyrighted works only under fair use or a licence. Platforms that fail to comply could lose their local business permits. The draft does not say how much human contribution is enough or how this might be policed.
The government’s sharper position. On 27 October the Director General of Intellectual Property, Hermansyah Siregar, set out the government’s proposals. Commercial text and data mining, including AI training, would need a reasonable licence, either individually or collectively through a collective management organisation (LMK). Non-commercial research by affiliated researchers would not infringe. Developers would have to publish summaries of their training data, and government material also proposes limits on imitating expressive style.
Beyond AI. Tirto’s reading of the draft is that permission is retained for publishing, copying, translating and distributing a work, while adaptation, performance, announcement and communication would be cleared by paying royalties through an LMK. Other reported provisions include a resale right for visual artists, a public lending right for authors, more transparent royalty governance, express protection for journalistic work, and a “dynamic injection” mechanism for blocking pirate sites faster.
The objections. At the 11 November hearing Google opposed a single centralised LMK issuing blanket licences. It argued that licensing all training data is unworkable given the number of web pages involved, proposed a text-and-data-mining exception with an opt-out such as robots.txt, and asked for safe-harbour protection to stay. TikTok asked that the AI-disclosure duty not apply where AI is merely an assistive tool. Meta warned that publishers could lose traffic, advertising and subscription revenue. One Indonesian commentator notes that Meta blocked news links in Canada after similar legislation.
On the other side, the Press Council and media organisations had already asked in February for the 2024 publisher-rights regulation to be raised to statute.
The regional backdrop. Japan and Singapore allow AI training through statutory text-and-data-mining exceptions, ironically written before the generative-AI boom. India’s government working paper of December 2025 proposed a mandatory licence with royalties paid through a central collective, Australia rejected a text-and-data-mining exception in favour of paid licensing, and Korea declined to add an exception in February 2026. These comparisons come largely from industry-body summaries (CCIA and BSA), which have their own agendas, but I’m reasonably confident that stands scrutiny.
The View From The Beach (Inferred)
Yeah, a numbered list. But yes, I do those sometimes, usually when there are distinct points to get across. It’s called grammar and formatting, and such designs existed before November 2022, so lay off with the “I know ChatGPT wrote every word” emails, please.
1. Indonesia is joining a licensing-led camp. Set beside India, Australia and Korea, the bill points towards paid licensing rather than the permissive Japanese and Singaporean model. None of these governments calls it a bloc, but a favourite word of mine, and the frame that makes the story matter beyond Jakarta.
2. News payment and AI training are being folded into one channel. Both run through state-supervised collective bodies. The press lobby supplied years of political momentum while AI supplied the urgency and the headline. My guess is the AI provisions are why a news-compensation campaign became a general copyright rewrite.
3. Two undefined phrases will decide the bill’s reach. A demanding “human involvement” test would put AI-assisted editorial, translation and design workflows at risk of falling outside protection; a light one changes sweet FA. A style-imitation ban is going to be almost impossible to draft in detail, because style is not normally protected by copyright, but it could be a game-changer if it in effect said “AIs may not follow prompts that ask it to create something in the style of any work itself under copyright. So okay for Shakespeare but not Ghibli.
4. Business-permit sanctions are leverage over firms with a local presence. Meta’s Canadian precedent suggests one possible response is withdrawing features rather than paying. Whether anyone would try that in a market of Indonesia’s size is an open question. With 280 million people, Indonesia is the fourth most populous country on the planet.
5. Mandatory licensing may not survive intact. Google’s opt-out proposal hands lawmakers an obvious compromise: a text-and-data-mining exception with a machine-readable reservation of rights, closer to the EU model, in place of compulsory licensing.
What we don’t know
Gaps in the public record as of 15 November 2026.
- The current text. The reporting rests on a draft reviewed in July and on government and company statements since. Changes made during committee deliberation may not be public. If they are, I missed them.
- The thresholds. How much human contribution counts, what “distinctive style” means, and what form AI disclosure takes, including whether tool-assisted use is exempt, as TikTok wants.
- The collective-licensing architecture. One LMK or several, who supervises them, what tariffs apply, and how money reaches individual authors and publishers rather than stopping at the institution.
- Reach. Whether obligations apply to models trained abroad but offered in Indonesia, and whether foreign rightsholders could collect.
- The legal basis of the “fair use” route. Reporting says training is permitted under fair use or licence, but not how fair use would operate in Indonesian law.
- The vote. A 12 December plenary is a target, not a schedule. Initiative bills can slip, and implementing regulations may follow any vote.
The View From The Beach Wrap-Up
Whatever the outcome, this is where the licensing-led approach gets stress-tested in a major emerging market, with the usual suspects lined up against it and the real design questions still open.
For publishers the big questions are structural rather than predictive: how “human involvement” is defined, who runs the collection bodies and whether royalties reach the people who made the work, and what training-data disclosure will actually require.
The 12 December target will tell us whether the bill moves forward or stalls. The final wording of those three provisions will tell us more about what matters. Preparation, as ever, beats prediction.
This post first appeared in the TNPS LinkedIn Analysis Newsletter.
