Australia’s official music charts are introducing a new eligibility rule for recordings made with generative AI. The Australian Recording Industry Association, or ARIA, has confirmed that wholly AI-generated tracks will no longer qualify for the ARIA Charts, while recordings that use generative AI in a supporting role can remain eligible.
The change takes effect for the ARIA Chart dated Monday, 31 August 2026, which will be published on Friday, 28 August. This is not a legal ban on AI music in Australia, and it does not force streaming services or radio stations to remove AI-generated tracks. It changes what can qualify for Australia’s main industry charts and, by extension, ARIA Awards eligibility.
More importantly, ARIA has drawn a practical line between AI-generated and AI-assisted music. That makes provenance, production records, and the role played by human performers much more important than simply asking whether AI was used at any point in the workflow.
What ARIA has actually banned
In its primary-source announcement on AI chart eligibility, ARIA says a recording developed using generative AI is eligible only if it is substantially human-made and raises no concerns about stream or chart manipulation.
ARIA is using the same definitions introduced through the music industry’s AI labelling standard in July. Under those definitions, a recording is AI-generated if generative AI created all or the primary portion of its creative elements. ARIA specifically gives an AI-generated lead vocal, an AI-generated key instrumental performance and an entirely prompt-generated track as examples.
An AI-assisted recording is different. The track must be created substantially by humans and express human creativity, with humans performing the lead vocal and primary instruments. Generative AI can still be used for some expressive elements without automatically making the recording ineligible.
| Production scenario | How ARIA’s published definition treats it | Chart position |
|---|---|---|
| A prompt generates the full song | AI-generated | Ineligible |
| A human writes or arranges the track, but AI generates the lead vocal | AI-generated because the lead vocal is a primary creative element | Ineligible |
| AI generates a key instrumental performance | AI-generated | Ineligible |
| Humans perform the lead vocal and primary instruments, with generative AI used for secondary expressive elements | AI-assisted, provided the recording remains substantially human-made | Potentially eligible |
This makes the headline narrower than a blanket “AI music ban”. A track can contain generative AI and still chart. The key question is whether AI supplied the main creative performance or assisted a recording whose core performance remains human.
The difficult part is proving where AI stopped being an assistant
The obvious enforcement problem is that modern music software increasingly mixes conventional editing, machine learning and generative features inside the same production environment. A producer may use AI for a texture, backing element, sound replacement or edit without thinking of the finished track as “AI music”. Another producer may generate most of a track and then make a few manual changes.
A recurring concern among musicians discussing the policy is exactly where the “supporting role” ends and primary creation begins. A simple percentage threshold would be easier to explain, but it would also be easy to game. Ten seconds of human performance does not necessarily make a prompt-generated song substantially human-made, while extensive AI-assisted processing does not necessarily replace the human performance at the centre of a recording.
ARIA’s approach therefore focuses on the role AI played in the creative elements, not merely on how many tools in the production chain used AI. That is a more defensible test, but it also means an audio detector alone cannot settle every dispute. A detector may flag characteristics in the finished file, but it cannot reconstruct who performed the lead vocal, which stem came from a generator, or how much of the arrangement existed before an AI tool was introduced.
Provenance is becoming part of chart eligibility
ARIA says it can decline an ineligible recording, remove it from the charts prospectively or retrospectively, adjust chart positions, withdraw accreditations and revoke or request the return of an ARIA number one award. An ineligible recording will also be ineligible for an ARIA Award. Artists and representatives can dispute an exclusion and provide evidence supporting the recording’s eligibility.
That creates a practical requirement not obvious from the headline: creators who use generative AI and still want chart eligibility should be able to demonstrate how the recording was made.
- Keep the original human vocal and instrumental takes.
- Retain project files, stems and version history rather than only the final master.
- Record which AI tools and model versions were used on which parts of the track.
- Keep prompts and generated assets where they materially affected the recording.
- Preserve contributor credits and production notes that identify human performances.
- Keep evidence that the services and source material used in the workflow were lawful and authorised.
ARIA has not published that list as a formal evidence checklist. It is simply the kind of production record that gives an artist something concrete to provide if eligibility is challenged. A text declaration saying “AI-assisted” is much weaker than a project history that shows human lead vocals, human primary instruments and the specific parts generated by AI.
Watermarks and AI labels will help, but they will not solve provenance
ARIA’s new chart policy is closely tied to the industry’s AI-Generated and AI-Assisted labelling system. That creates a useful route for platforms, distributors, and chart bodies to carry the creator’s declaration through metadata, rather than trying to infer everything from the final audio file.
Labels are still only one part of the answer. A label can be missing, incorrect or disputed. A watermark can be removed or lost during processing. Neither one proves how every stem was created. The stronger long-term model is likely to combine labels with production provenance: who created a component, which system generated it, what licence applied and whether the human contribution remained primary.
There is also an important scope limit. ARIA’s July labelling framework covered generative AI use in sound recordings and said it did not yet cover lyrics, composition, music videos or cover art. That means a human-written lyric sheet alone does not appear to rescue a recording in which the lead vocal or a key instrumental performance was generated by AI. The chart test is focused on the recording and its primary creative elements.
What should AI music creators change now?
For creators who have no interest in chart eligibility, very little changes immediately. They can still release AI-generated music through services that accept it, subject to those services’ own policies and the usual copyright and licensing questions.
For artists who do want chart or ARIA Award eligibility, the production decision now needs to be made earlier. If the lead vocal or a primary instrumental performance is generated by AI, ARIA’s published definition indicates that the recording should be treated as AI-generated. Adding human edits later should not be assumed to convert it into an AI-assisted track.
This is especially relevant to creators experimenting with AI singing voice generators for music. A generated lead singer is not a minor production aid under ARIA’s definition. The lead vocal is one of the examples ARIA expressly uses to identify an AI-generated recording.
Creators can also use our AI Music Generator to understand how quickly prompt-led generation can move from assistance to primary creation. For experimentation, that is useful. For a chart-eligible release, the safer workflow is to establish the human performance first, then use generative tools around it rather than asking the generator to supply the main performance.
AI music products now need better provenance features
The policy is also a product design problem for AI music companies. Generators have spent the past few years competing on output quality, speed and control. Chart and platform rules create a new category of feature: evidence.
A useful export could record whether a stem was generated, transformed or human-recorded, along with model identifiers, timestamps, licence information and contributor credits. A signed generation receipt or tamper-evident project log would be more useful in a dispute than a generic “made with AI” badge.
Products should also separate genuinely assistive tools from primary generation in their interfaces. Noise removal, timing correction and mastering assistance do not present the same authorship question as generating the singer or the central instrumental performance. Treating every AI feature as equivalent makes compliance harder for users and encourages vague disclosures.
This is an industry eligibility rule, not an Australian AI music law
For developers and creators outside the music industry, the scope is worth keeping clear. ARIA is not outlawing AI-generated songs, and the rule does not prevent Australians from listening to them. It changes eligibility for an industry chart and associated awards.
ARIA has also called on radio and other organisations involved in promoting music to consider similar changes, but those organisations would need to adopt their own policies. The immediate effect is therefore narrower than a nationwide platform ban.
The broader signal is more significant. ARIA has aligned its charts with global industry principles from IFPI, so the practical boundary between AI-generated and AI-assisted music now has a real commercial consequence. If other charts, awards programmes, distributors or radio networks adopt comparable definitions, provenance could become a normal part of releasing music rather than something only discussed after a dispute.
The real change is that “made with AI” is no longer specific enough
ARIA’s rule does not eliminate AI from commercial music. It forces creators and platforms to answer a more useful question: what did the AI actually create?
If AI generated the lead vocal, the key instrumental performance or most of the recording, the track falls on the AI-generated side of ARIA’s line and loses chart eligibility. If humans remain the primary creators and performers while generative AI plays a supporting role, the track can still qualify, provided it is substantially human-made and raises no concerns about manipulation.
For creators, that turns production records into evidence. For AI music companies, it turns provenance and metadata into product features. For chart bodies, it creates an enforcement problem that cannot be solved simply by running finished songs through an AI detector. The policy is likely to be most effective where those three pieces work together.
