Indonesia AI Copyright Bill Is Not the Lock | Elacity
Indonesia's AI copyright bill would require a reasonable license for commercial AI training. A law can demand a license. It cannot, by itself, make the work carry one.
Indonesia's AI Copyright Bill Would Require a License. The Work Still Wouldn't Carry It.
If you publish work that AI companies want to train on, Indonesia's AI copyright bill would require them to obtain a license before they use it for commercial training. That is more than an opt-out form can give you. It is still not a license that travels with the work itself.
The bill is not law yet. The legislature and the government are still debating it, and they have set a target of taking it to a plenary session before December 12, according to ANTARA. Until that vote happens and the text is enacted, every provision below is a proposal, not a rule you can rely on.
What the Indonesia AI Copyright Bill Actually Proposes
Indonesia's intellectual property chief, Hermansyah Siregar, set out the government's proposal on October 27. Text-and-data mining for noncommercial scientific research, done by researchers affiliated with a research institution or a university, would not be treated as copyright infringement. Commercial text-and-data mining, including the training of AI models, would have to carry a "reasonable license," granted either individually or collectively through a collective management organization.
The same proposal would require a published summary of the data used to train an AI model. A work produced by AI with no human intellectual involvement would not receive copyright protection. Both points are still in discussion, not in force.
A noncommercial research exception is a fair line to draw, and a commercial license requirement is more than most opt-out regimes offer. We have already set out where opting out of AI training stops: the request lives in a policy the company can ignore, change or never see. A statute that says the company needs a license is stronger than a tag in a robots.txt file. It is also still a rule about what should have been agreed before the copy was made.
At the Hearing, the License Mandate Was Treated as Unworkable
On November 11, at a parliamentary hearing, Google and Meta asked for a clear text-and-data mining exception that would cover commercial AI training on material accessed lawfully, Katadata reported. Meta's public policy head in Indonesia, Berni Moestafa, said draft articles could be read as requiring separate permission for each work used to train a commercial model, which he argued would be hard to implement at the scale of training sets. Google's government affairs manager in Indonesia, Agung Pamungkas, called a technical licensing obligation unworkable given the scale of the indexed web.
That is a company position, not a measurement of what creators are owed. The practical problem they are pointing at is real: no individual writer can sit across the table from a lab and license each file. That is exactly why collecting societies exist. It is also why the person who wrote the work often never holds the meter.
The same hearing had a second fight running in parallel. Asia IP reported that Google and Meta also opposed proposals that would require platforms to pay publishers for displaying news links, and that TikTok asked lawmakers not to apply AI disclosure rules to assistive use and to keep safe-harbor protections, arguing that platforms cannot reliably determine who owns each upload. Those are real design problems for a platform. They are not an argument that your work should be free to train on.
A License in a Statute Does Not Travel With the Work
Even if the bill passes in the form the government has proposed, the license would live in Indonesian law and, for many creators, in a collective management organization. The file you posted, the song you sold, the essay you published would still move as an ordinary copy. The next company, in the next country, that does not face this statute would be back to taking first and arguing later.
A published summary of training data would help you see whether your work was in the pile. Seeing the pile is not the same as controlling the use. Transparency after the fact still starts from a copy that has already been made.
That is the same shape we described when we asked who holds the meter in AI content licensing models. A "reasonable license" defined in a statute, then split by a collective, still leaves someone else to decide what reasonable means, someone else to count the use, and someone else to send whatever remains. This is a question of Ecosystem & Governance. The gap is mechanical. A law can demand that a license exist. It cannot, by itself, attach that license to the work so that use without it fails.
The License Has to Live in the Work
That is what we are building Elacity dDRM for. In the design, you package the work into a Wealth Capsule: a programmable, royalty-bearing good with the terms you set written in, so that use is checked against those terms rather than against a later policy or a collecting society's formula. The point is not a promised payout. The point is that you own the asset, you set the terms, and those terms are designed to travel with the work.
A national license mandate, even a good one, is not the last step. Indonesia can require a reasonable license for commercial training. You still need a way to grant one that has to be met at the point of use, not only in a statute the next jurisdiction may never write.
If you want more on how rights can travel with the work, Follow Elacity on X.