Distributors check the files. Platforms check the plays. Parla will check before they pay.

We've been working on this for a long time, and we are finally ready to roll out Parla. Parla works out what matters in a given situation in milliseconds.

What Parla does

Parla (Profile Associative Relevance Layer) helps an AI agent find the right record at the right moment.
Before the AI agent recommends a song, Parla points it to the record showing whether the artist was around before that song appeared.
The record is the proof, and Parla makes sure the AI agent sees it.

A fraud built to stay small

Michael Smith pleaded guilty in March 2026 to collecting more than $8 million in streaming royalties. He used hundreds of thousands of AI-generated songs and bot accounts that played them over and over. Prosecutors say he spread the plays across thousands of songs because a few songs with billions of plays would have been caught. He is due to be sentenced on 6 October, and prosecutors are asking for at least 46 months.

The money came from the shared royalty pool. Streaming services divide one pot of money among everything that is played, so every fake play takes a tiny amount from every real artist. No single artist notices.

Don't the platforms already handle this?

They handle part of it.
Qobuz reports that 60 percent of the plays on tracks it identifies as AI-generated are fraudulent, and it keeps them out of royalty payments.
Deezer reports 85 percent.
These are plays that were watched, judged fake, and removed.

1. That is one question, whether a person really listened.
2. A second question gets no check at all, whether the artist and the song existed before the release.

Distributors check the files they receive.
Streaming services accept what distributors deliver and monitor how people listen. Smith designed his plays to look ordinary, and he did not need to make his artists look real because nobody asked.

ProofProfile answers the second question. The platform's tools do something different and remain necessary.

ProofProfile

ProofProfile is a register of 2,006,172 artists and 9,900,355 recordings, taken from music databases as they existed before 2022. Each recording is registered and carries a Bitcoin timestamp, a date stamp that cannot be changed afterward. Anyone can verify that a recording is exactly as it was on the day it was stamped, without needing ProofProfile's permission.

An artist who is in the register was on record before the release. An artist who is missing is simply unknown. That does not mean fraud, because every new artist is missing at first. It means someone should check before money moves.
ProofProfile does not detect fake plays or AI-generated audio. It does not identify who wrote a song, who owns it, or whether a human made it. It moves no money. It shows what existed and when, and the paying company decides what to do with that.

Why the AI agent needs Parla

Every day, more AI agents take actions on people's behalf. On 29 September, OpenAI said ChatGPT has 1.2 billion weekly users and launched always-on AI agents that operate independently. Making a convincing fake artist now costs almost nothing, since you can generate the name, voice, cover art, and life story in an afternoon. An invented artist can reach far more people through an AI assistant than through any website.

An AI agent about to make a payment sees a name, a title, and a code. None of them looks like the record that would show whether the artist is genuine. Parla attaches short instructions to ProofProfile records that specify when each applies. One says to read the record when a release arrives under a registered artist's name. Another says to read it when a payment names a specific recording.
The instructions are not proof.
They feed the proof to the AI agent, and the proof stays in ProofProfile.
Parla's service searches the register and tells apart artists who share a name.

How to judge it

The test is simple. Can the AI agent tell two artists with the same name apart? Does it pick the right song? The register behind it covers 2,006,172 artists and 9,900,355 songs, all certified as human-origin, and the test runs across all of them.

Fraudsters adjust to whatever limit a platform sets. A dated record of who was already there depends on no limit, and nobody can create one after the fact.