ARIA and Beatport both draw an AI line. Only one lets a human hear the appeal

Two gatekeepers drew a line on AI-made music in August. On 25 Aug, ARIA said that from the chart dated 31 Aug, "wholly AI-generated tracks" can't enter the ARIA Charts, while recordings that use generative AI in a "supporting role" stay eligible. Two weeks earlier, on 12 Aug, Beatport said it will withhold fully or majority AI-generated tracks when they're uploaded, and tag AI-assisted ones it accepts.

The headlines read the same: no fully machine-made music, AI as a tool is fine. The way each one decides is different, and for artists that is the part that matters.

What triggered the ARIA rule?

A chart hit. ABC News reports that Queensland DJ and producer Josh Fawaz's cover of Madonna's "Like a Prayer" reached No. 4 on two ARIA charts in July and became the most-played song on Australian radio. Fawaz said he used AI "as a tool". He later updated the Spotify listing to say the vocals and drums were AI-generated. An ARIA spokesperson told the ABC that the IFPI principles would likely disqualify music with AI-generated lead vocals or key instrumentals.

Those principles come from the global recording industry body. On 30 July IFPI said recordings made with generative AI shouldn't be included in official charts where there is "reason to believe" they aren't "substantially human made", raise manipulation concerns, or used an AI service that isn't "properly authorised and lawful". It named ARIA among the national charts it was working with to apply them.

How does ARIA decide what counts?

By declaration, with examples. ARIA's FAQ says every release submitted to its survey now needs a declaration about generative AI use, and ARIA works from that. It sorts recordings into three groups:

• AI-generated (not eligible): AI produced the recording or the primary part of its creative elements, including tracks made entirely from a prompt. An AI lead vocal, or any AI key instrumental part, puts a track here.
• AI-assisted (eligible): humans wrote the song and performed the lead vocal and primary instruments. A human lead with AI backing vocals underneath is the FAQ's own example.
• AI in production (eligible): uses such as AI mastering, stem separation and AI drum machines.

If someone raises a credible concern, ARIA says it will contact the rights holder, explain the concern and let them respond. A disputed track can go to the ARIA Chart & Marketing Committee and then the ARIA Board, which makes the final call. ARIA can drop a recording from the chart, retrospectively too, and revoke an ARIA #1 Award.

One notable gap: ARIA has not applied IFPI's "authorised and lawful" test to chart eligibility yet, saying licensing between music companies and AI firms is still "a rapidly developing area". The FAQ does require that recordings comply with copyright law. So a track built on an unlicensed model isn't excluded for that reason alone.

How does Beatport decide?

By detection. Beatport expanded its partnership with Beatdapp, a company it already used for streaming fraud, to cover AI detection. Beatport says Beatdapp's technology can "distinguish between fully generated and AI-assisted tracks". Fully or majority AI tracks are withheld at ingestion and the rights holder is told directly. AI-assisted tracks are accepted if the finished record "remains majority human-made", and are tagged so Beatport's curators can see what they're featuring. NME quotes CEO Matt Gralen drawing a line between a tool that assists human creation and a system that replaces it.

Beatport backs the move with its own DJ survey: it says 60% of respondents wouldn't play AI-generated tracks in their sets. That is Beatport's figure about Beatport's users, not an independent poll.

Beatport's announcement doesn't define "majority", doesn't give examples like ARIA's, and doesn't describe any appeal or review if a human-made track gets withheld.

Who carries the risk?

The artist, in both cases. But how much risk differs.

Under ARIA, the artist's own declaration is the main input. That puts honesty on the artist, but it also leaves the description of the work with the person who made it, and a wrong call can be argued in front of people. Under Beatport, a classifier makes the first decision about whether a track goes on sale. As Dizko argues, authorship is about how a record was made, but a detector only hears the finished audio. A carefully played live take that was recorded cleanly and mastered by an automated service may read as suspiciously tidy, and a generated track that was resampled and re-recorded may pass. Every classifier needs a threshold, and the false positives land on human producers who have no easy way to prove their process.

Dizko also points out that machines already sit at many steps of modern production, from stem separation and automated mastering to generated drum variations, which are much the same tools ARIA lists as fine. For the electronic producers Beatport serves, a vague "majority" test meets a lot of mixed workflows. We couldn't find much public evidence yet of tracks wrongly withheld, so treat that as a risk, not a documented pattern.

What we think

Drawing a "human-led" line is defensible, and both rules are better than none. The problem is that the word doing the work ("supporting" for ARIA, "majority" for Beatport) is vague, and the real policy is whoever applies it.

ARIA has done the less glamorous, more useful thing. It published examples a producer can check before release, built the rule on a declaration the artist controls, and wrote down a disputes route that ends with people. That costs ARIA more effort per case, but it treats artists as able to describe their own work.

Beatport has handed the first judgement to a detection vendor. The second-order effect is that Beatdapp's threshold becomes a commercial gate for every label and producer selling on Beatport, without the definitions or appeal process published alongside it. Labels will start asking for provenance on every submission. Producers who use AI for one element will have an incentive to say nothing and hope the detector agrees, which is the opposite of the transparency Beatport says it wants.

Two fixes would close most of the gap. First, let artists disclose AI use on their own terms at upload, using the same AI-generated and AI-assisted labels ARIA now relies on, and weigh that declaration before the detector's score. Second, no release should be withheld on a detector's say-so alone: give a named human review, a deadline and a way to submit stems or session files as evidence. ARIA has shown that is workable. Beatport, and any store or distributor that adds detection, should match it.