The Australian Recording Industry Association updated its Chart Code of Practice so that a recording generated by AI is no longer eligible for the ARIA Charts. The change took effect with the chart dated Monday 31 August 2026, published on Friday 28 August. A recording made by humans that used generative AI stays eligible, and ARIA describes that category as one where human creative input is the primary driver of the work. The association gave a worked example: a lead vocal generated by AI makes a track ineligible, as does any key instrumental component, while a human lead vocal with AI backing vocals underneath counts as AI-assisted and can chart. Where ARIA decides a recording is ineligible it can refuse it for survey, remove it from the charts prospectively or retrospectively, adjust chart positions, withdraw accreditations and revoke an ARIA number one award. Artists can contest an exclusion and supply evidence through a disputes process the same update rewrote.
The chart does not inspect the record
ARIA is not proposing to run audio through a classifier. Its guidance puts the obligation somewhere else, saying that what matters for the charts is that the declaration made to ARIA about the use of generative AI is accurate. To separate a generated recording from an assisted one, it applies the definitions in the labelling standard the global music industry announced on 10 July 2026, which ARIA had already welcomed.
So the rule is a paperwork rule. That is not a criticism by itself, because almost every chart rule is a paperwork rule, and eligibility for release date, format and price has always run on what a label tells the chart compiler. It does change what the announcement means. ARIA has not acquired a way to detect AI in a recording. It has acquired a ground on which to act when a declaration turns out to be wrong, and a process through which an artist can defend one.
The enforcement powers are unusually broad for a chart, and the retrospective ones are the interesting part. A chart position is a historical record that other things get built on, including radio programming, accreditations and the number one award ARIA can now ask to have returned. Reaching back into a published ranking is a heavier act than declining to publish one.
The line is finer than the disclosure
The distinction ARIA drew is precise. An AI lead vocal makes a recording ineligible and AI backing vocals under a human lead do not. Nothing published on a streaming service currently records that difference.
The track that put this question in front of the Australian industry shows the gap. Josh Fawaz’s cover of Madonna’s Like a Prayer spent four weeks at number one on the national airplay chart and has passed 39 million Spotify streams. In July, after criticism of how the record was made, Fawaz added AI credits to it on Spotify, listing generative AI vocals and AI drums. Those credits are a real disclosure and more than most releases carry. They also do not say whether the generative vocals sit on the lead or underneath it, which is the exact fact ARIA’s test turns on.
None of that establishes anything about the track’s eligibility, and we are not suggesting it does. The rule applies from the chart dated 31 August, ARIA has said the change gives everyone releasing music in Australia a clear rule to work to rather than singling out a release, and Fawaz has defended AI as a tool while calling the new rules a great move. The point is narrower and it survives all of that. A public credit field one level coarser than the eligibility test cannot answer the eligibility question, so the answer has to come from a declaration the public does not see.
A chart is a scoreboard and not a door
Beatport screens at upload, which is a gate: a rejected track is not in the store. Spotify’s badge attaches to a profile and changes how an account is described and recommended. A chart holds neither position. ARIA surveys sales, streams and airplay generated by shops, services and radio stations it does not operate, and it publishes a ranking of what happened.
Removing a recording from that ranking does not remove a play. Like a Prayer reached number four on the ARIA Australian singles chart after peaking at number two, topped the Australian dance singles chart, and did that on the back of airplay decided by programmers at commercial radio. Those programmers are the mechanism, and ARIA’s statement addresses them by asking everyone with a role in deciding what gets played and promoted to Australian audiences to support human artistry. That is a request. The chart rule is the only part with teeth, and it operates on the scoreboard rather than on the thing being scored.
Chief executive Annabelle Herd put the reasoning plainly, drawing a distinction between artists using AI tools in their work and music generated wholesale by services built on artists’ recordings. The second is the target. The rule reaches it through eligibility because eligibility is the instrument a chart compiler holds.
Three institutions, three different points in the chain
This is the third line drawn in a month by an organisation that sits at a different place in the same chain. Beatport screens fully AI-generated tracks at upload and keeps AI-assisted work, on a threshold it has not defined. Spotify badges profiles whose identity is generated, which describes the name rather than the record. ARIA now rules on whether a recording may appear in a ranking, on a definition borrowed from an industry labelling standard.
What the three share is that none of them listens to the audio to decide. Each reads something a person supplied, whether that is an upload declaration, an account identity or a chart submission. The classifier work that does exist sits further upstream, and when a service applied one to material reaching it, it found signs of AI in 38.5 percent of the tracks it analysed, which were pitches rather than released records. The distance between those two populations is the whole problem, and a declaration is currently the bridge across it.
What would make the rule bite
The labelling standard announced on 10 July is doing more work here than the chart rule is, because it is the thing that decides what a declaration means. If its tags travel with a recording through distributors, stores and streaming services, then ARIA is applying a definition that everyone in the chain already had to answer. If they do not travel, ARIA is asking a question at the end of a pipeline that did not record the answer anywhere else.
Two things are worth watching over the next few months. The first is whether ARIA ever uses the retrospective powers, since a rule enforced only prospectively is a much smaller rule than the announcement suggested. The second is whether the credits that services display start to distinguish a generated lead from generated backing. Until they do, the most detailed AI disclosure a listener can see on a major release will remain too coarse to answer the question the chart now asks.