Indonesia’s draft copyright bill could protect AI-assisted works, but it still does not say how much human creativity is enough.
That missing line is the central problem in the human involvement test for AI copyright. A creator may spend hours directing, selecting, rewriting, arranging, and editing machine output, yet the draft reported in July 2026 provides no published threshold for deciding when that labour becomes authorship.
The wider Indonesia’s proposed AI copyright rules would exclude fully machine-generated output while allowing copyright protection for AI-assisted works that show sufficient human involvement. The distinction looks tidy on paper and gets messy fast when a project mixes prompts, source material, generated passages, manual edits, and dozens of rejected versions.
Indonesia has not yet published a binding formula for how much human input AI copyright requires. The bill remains a draft, so creators cannot treat its present wording as a finished registration standard.
A workable interpretation would probably examine control over expressive elements rather than effort alone. Spending ten hours generating images does not automatically prove authorship when the system decided the composition, lighting, faces, textures, and final details.
That difference matters for copyright protection for mixed human and AI works. A photographer who takes an original photograph and uses AI only to remove an unwanted object has a much clearer human-authorship story than someone who enters a short prompt and accepts the first generated image.
The same issue reaches copyright ownership of AI-assisted music. A musician who writes the melody, records the performance, and uses AI for noise removal stands on firmer ground than a user who asks a generator for a complete song and merely chooses the preferred result.
A prompt can request a mood, framing, subject, genre, or pacing while the model still determines the exact words, shapes, notes, and visual relationships that appear. Under a human-control test, that gap makes prompt-only authorship difficult to defend.
Repeated prompting may also prove labour without proving authorship. For creators considering whether AI-generated art can be copyrighted after editing, the stronger issue is whether the editing added original expression that can be identified apart from the machine’s contribution.
Minor cleanup is unlikely to tell a convincing creative story on its own. Cropping an image, fixing spelling, changing a file format, or choosing one result from several may show judgment, but the human contribution could still be too thin to define the final work.
Substantial rewriting, hand-drawn additions, original performance, deliberate sequencing, and detailed visual compositing present a stronger case. Those acts move beyond asking for content and begin shaping the expression that the audience actually receives.
This kind of documenting human contribution to AI content is not glamorous, but it can separate a serious ownership claim from a vague statement that the creator worked hard on the prompt. It may also help businesses identify which rights they can safely license, sell, or enforce.
The evidence should connect human choices to the final expression. For copyright protection for edited AI images, that could mean keeping the original generation beside the later composition, retouching, masking, typography, colour work, and hand-drawn additions.
For text, AI-generated work registration evidence may include tracked changes showing substantial rewriting, a human-created structure, original reporting, and independently written passages. Merely correcting spelling or selecting one output from several would present a weaker record of authorship.
The safest reading of the draft is that AI can remain a tool without becoming the legal creator. Until lawmakers define the threshold, creators relying on copyright protection for AI-assisted works will need to show where the machine stopped making expressive choices and where the human author clearly took over.
That missing line is the central problem in the human involvement test for AI copyright. A creator may spend hours directing, selecting, rewriting, arranging, and editing machine output, yet the draft reported in July 2026 provides no published threshold for deciding when that labour becomes authorship.
The wider Indonesia’s proposed AI copyright rules would exclude fully machine-generated output while allowing copyright protection for AI-assisted works that show sufficient human involvement. The distinction looks tidy on paper and gets messy fast when a project mixes prompts, source material, generated passages, manual edits, and dozens of rejected versions.
Human involvement in AI-generated works stays undefined
The uncertainty affects anyone asking who owns AI-generated content in Indonesia. The answer may depend less on which tool was used and more on whether the final work carries identifiable creative decisions made by a person.Indonesia has not yet published a binding formula for how much human input AI copyright requires. The bill remains a draft, so creators cannot treat its present wording as a finished registration standard.
A workable interpretation would probably examine control over expressive elements rather than effort alone. Spending ten hours generating images does not automatically prove authorship when the system decided the composition, lighting, faces, textures, and final details.
That difference matters for copyright protection for mixed human and AI works. A photographer who takes an original photograph and uses AI only to remove an unwanted object has a much clearer human-authorship story than someone who enters a short prompt and accepts the first generated image.
The same issue reaches copyright ownership of AI-assisted music. A musician who writes the melody, records the performance, and uses AI for noise removal stands on firmer ground than a user who asks a generator for a complete song and merely chooses the preferred result.
AI prompts alone create a weak copyright claim
Detailed instructions can show intention, taste, and persistence, but they may not show control over the final expression. That is why copyright protection for detailed AI prompts should not be confused with copyright in the resulting image, song, video, or article.A prompt can request a mood, framing, subject, genre, or pacing while the model still determines the exact words, shapes, notes, and visual relationships that appear. Under a human-control test, that gap makes prompt-only authorship difficult to defend.
Repeated prompting may also prove labour without proving authorship. For creators considering whether AI-generated art can be copyrighted after editing, the stronger issue is whether the editing added original expression that can be identified apart from the machine’s contribution.
Minor cleanup is unlikely to tell a convincing creative story on its own. Cropping an image, fixing spelling, changing a file format, or choosing one result from several may show judgment, but the human contribution could still be too thin to define the final work.
Substantial rewriting, hand-drawn additions, original performance, deliberate sequencing, and detailed visual compositing present a stronger case. Those acts move beyond asking for content and begin shaping the expression that the audience actually receives.
Proving creative control over AI output takes evidence
That control will be easier to prove when the process leaves evidence. Creators should preserve original drafts, source files, project histories, layered edits, version logs, and records showing which elements were written, drawn, filmed, performed, or arranged by a person.This kind of documenting human contribution to AI content is not glamorous, but it can separate a serious ownership claim from a vague statement that the creator worked hard on the prompt. It may also help businesses identify which rights they can safely license, sell, or enforce.
The evidence should connect human choices to the final expression. For copyright protection for edited AI images, that could mean keeping the original generation beside the later composition, retouching, masking, typography, colour work, and hand-drawn additions.
For text, AI-generated work registration evidence may include tracked changes showing substantial rewriting, a human-created structure, original reporting, and independently written passages. Merely correcting spelling or selecting one output from several would present a weaker record of authorship.
The safest reading of the draft is that AI can remain a tool without becoming the legal creator. Until lawmakers define the threshold, creators relying on copyright protection for AI-assisted works will need to show where the machine stopped making expressive choices and where the human author clearly took over.
