Anghami's COO says AI detection tools cannot read Arabic music reliably, right as Spotify starts labelling

Spotify said on August 11 that from mid-September it will start putting AI Persona badges on artist profiles whose identities look artificially generated. A week later the chief operating officer of the biggest streaming platform in the Middle East and North Africa pointed at a problem underneath that plan, which is that the tools doing the deciding were never built to read his region’s music.

“

Global AI tools, both for creation and detection, aren't really built with Arabic music in mind. They're trained and benchmarked on Western catalogues, meaning they don't serve Arabic-language artists as well as they should, and they don't identify Arabic music as reliably either.

Al Ogaily made the comments to WIRED Middle East, and the timing is what gives them weight. A detection gap is an accuracy problem while nobody acts on the output. It becomes a distribution problem the moment a platform starts labelling.

Why the Spotify badge turns a measurement gap into lost reach

Spotify’s badge is narrower than most coverage suggests. It targets the public identity a profile presents, meaning the name and the images, rather than how the music itself was made, and artists can disclose AI in the creative process separately through AI Credits and SongDNA. Artists can self-declare, and those labelled can confirm or appeal.

The consequence is where it bites. Profiles identified as AI Personas are excluded by default from Spotify’s editorial and algorithmic recommendations, including Discover Weekly and Release Radar. So the badge is not a note on a page, it is a switch on discovery.

Now put Al Ogaily’s point next to that. If the systems assessing profiles carry the same Western training bias he describes in creation and detection tools, then the error rate for Arabic-language artists is both higher and more expensive, because the penalty for a wrong call is reach.

What Anghami has been saying about this since March 2026

This is not a new position for the company. In March, Anghami integrated Cyanite’s audio-based tagging across 2.5 million songs, and the same structural gap sat underneath that deal. Arabic music uses the Maqam system with 24 quarter tones where Western theory works in 12 semitones, and models trained on Western material tend to round those microtones to the nearest Western pitch. The Missing Melodies study put Middle Eastern music at 0.4% of major training datasets.

Al Ogaily is now applying that same argument to detection, which is the harder version of the problem. A tagger that misreads a maqam produces a bad genre tag. A detector that misreads one produces a verdict a platform acts on.

Anghami is dealing with the volume side too. Monthly song submissions have almost doubled since 2023, with much of that growth coming from AI-generated music, so the platform needs detection to work and is saying out loud that the available tools do not work well enough on its catalogue.

For artists working outside the Anglo-American mainstream, the practical takeaway is that the labelling era is arriving with error rates nobody has measured for your music. That makes the appeal process, and whether platforms publish per-market accuracy at all, the thing worth watching over the next two months rather than the badge itself.