No-AI Act & Evidence
Our work is made by people.
We write, play, arrange, record, draw, design, edit, mix, master, mess with things, fix things and occasionally break things. Our work includes music, logos, cover artwork, tour artwork and other visual material. It can involve many people, layers, effects, digital tools, strange production decisions and questionable ideas.
Whatever the starting point, the finished work is shaped by people working together – rehearsing, performing, improvising, arranging, recording and refining it.
None of that makes the result AI-generated.
Automated AI-detection systems are probabilistic classifiers. They can produce a forecast about a finished piece of work. They cannot establish who created it, how it was created, or what happened during the creative process that produced it.
We will not accept our music, artwork or other creative work being publicly labelled, presented, restricted or otherwise treated as AI-generated solely because an automated system classified it that way.
Any platform that publicly labels any of our work as AI-generated based solely on an automated detector, its confidence score, or a human review of that detector's output will lose our work. Where applicable, that means removing our entire catalogue from that platform.
We are not required to prove that our work is human-made because somebody else's system guessed otherwise. An algorithmic forecast is not a valid basis for publicly making claims about the provenance of our work. Platforms that choose to publish such claims are responsible for the consequences and the trail those claims leave behind – even if the label is later corrected or removed. That is precisely why an automated forecast should never be turned into a public claim in the first place.
There is no “we'll correct it afterwards” exception to this policy. Once a false claim about our work has been presented to an audience as fact, the damage has already been done. Removing the label afterwards does not undo that.
This is not a hypothetical policy. We left Facebook after our work was given an “AI generated” label while the qualification that it only may have been AI-generated was hidden behind an additional interaction. The work in question was not AI-generated. It was our work, made by us. That false label and the way it was presented are exactly why we left.
Read the according blog here →
And we will continue leaving platforms that do this. We would rather end up with little or no presence on major online platforms than accept having the provenance of our work publicly rewritten by an algorithm. Reach, streams, followers and convenience are not worth allowing an automated forecast to be presented as fact about something we created.
If necessary, we will distribute and stream our work through infrastructure we control instead.
A prominent “AI generated” label is not made acceptable by hiding “may have been generated”, a confidence score, disclaimer or other qualification behind a click, tooltip, information screen or other less-visible element.
Nor does displaying that uncertainty prominently solve the underlying problem. “98% likely AI-generated” is still an algorithm publicly making a claim about the provenance of somebody else's work. A probability does not become proof merely because the number is high.
If an automated system flags something we made, that result should trigger private investigation, not a public accusation. A platform can investigate whatever concerns it has. What we will not accept is an algorithmic forecast being turned into a public statement about our authorship nor provenance.
This is not merely a technical argument about classifiers.
There are real artists behind this work.
People can spend months or years writing, rehearsing, recording, arranging, drawing, designing and refining something.
Falsely presenting that work as AI-generated undermines the authorship of the people who created it and dismisses the human work behind it.
This can and will hurt the people behind that work. The damage begins the moment you make the accusation public—not when somebody finally notices your mistake and offers to correct it. You cannot simply undo that harm by changing a label afterwards.
Big corporations deploying these systems should seriously consider that before allowing an algorithmic forecast to make public claims about real people's work.
The consequences can also be financial. Using an automated AI classification to demonetize work, withhold revenue, restrict distribution or otherwise financially penalize an artist means making decisions about real people's income based on an algorithmic forecast.
An algorithmic forecast should never, by itself, become the basis for a financial penalty. Having a human review or approve the algorithm's output does not turn that output into proof. If a platform intends to withhold revenue, restrict distribution or otherwise penalize an artist over alleged AI generation, that allegation must be supported by independent evidence establishing the actual provenance of the work – not merely by an automated detector and somebody agreeing with it.
“A human reviewed the result” is not the same thing as “a human established what actually happened.”
If somebody is genuinely curious about how something we've made came to exist, they are welcome to ask us. We may share the story of how it was written, recorded, designed or otherwise created, and many of those stories are already or will become part of our website.
That is an invitation to learn about our work, not an obligation for us to prove that we made it.
A weather forecast can predict sunshine while you're suffocating in the rain. Calling an algorithmic forecast “AI generated” does not make it a fact. Taking an artist's income away based on that forecast does not make it more accurate.
We consider automated AI detectors fundamentally unsuitable as arbiters of creative provenance. When they falsely label human-made work as AI-generated, that is not merely a technical error but an insult to the people who actually wrote, performed, recorded, engineered, mixed, mastered, drew or designed it. A detector may provide a signal worth privately investigating, but it cannot establish how a work was created. Any architecture intended to prevent AI misuse is failing its purpose if it treats legitimate human creators as acceptable false positives, publicly accuses them, or penalizes them on the basis of an algorithmic forecast.
And for anyone who believes an algorithm's assumption should outweigh the artists who actually made the work:
Fuck off. We don't want anything to do with you.
Bad People Make Good Music