The companies that win the next decade of AI won't be the ones with the most data. They'll be the ones with the most defensible data.
There's a pattern in technology that repeats every time a new medium learns to copy things at scale. Someone builds a brilliant piece of infrastructure on top of other people's work, the value is undeniable, the users love it — and then the law arrives. It happened to music. It is happening, right now, to AI. The only open question is whether this generation of builders learns the lesson the first time or pays for it in court.
At Ché, a Cortex Research Group project, we've built our entire thesis around getting on the right side of that line before it's drawn for us. This is the argument for why ethically-sourced, attributed, licensed data isn't a compliance checkbox — it's the foundation the durable AI companies will be built on.
The Napster lesson
In 1999, Napster did something genuinely revolutionary. It made the world's music instantly available, peer-to-peer, for free. Tens of millions of people used it. By any product metric, it was a phenomenon.
It was also built entirely on copyrighted work that nobody had licensed.
When the record labels and artists sued, the courts didn't care that Napster itself wasn't the one uploading the songs — its users were. In A&M Records, Inc. v. Napster, Inc. (2001), the Ninth Circuit found Napster liable for contributory and vicarious copyright infringement: it knew its users were trading copyrighted files, it materially contributed to that infringement by providing the infrastructure, and it had the right and ability to police the behavior while profiting from it. Napster was ordered to filter out infringing material — a task its architecture couldn't actually do — and the service collapsed.
The key holding is the one every AI company should have tattooed somewhere: you can be held liable for infringement built on top of your platform, even when the infringing act was technically committed by someone else, if you knew about it, enabled it, and profited from it. Grokster reaffirmed it in 2005. The cases that followed only sharpened the principle.
Now reread that sentence and replace "users trading files" with "a model trained on a scraped corpus." The shape is identical.
AI is having its Napster moment
The most powerful models on earth were trained, in large part, on data their builders did not license. Books, art, code, journalism, music, photography — scraped at planetary scale, with no consent and no compensation to the people who made it. For a few years that was treated as a clever growth hack. It is now a pile of litigation.
The lawsuits are no longer hypothetical. Authors, artists, news organizations, music publishers, and image libraries have all brought copyright actions against major AI labs. The legal questions are still being fought — fair use is a real and serious defense, and AI training is not the same act as Napster's file-distribution. But the strategic risk is already settled, regardless of how any single case lands:
- The supply was never clean. A model trained on scraped data has no provenance. You cannot prove where any given datapoint came from, whether its creator consented, or that you have the right to use it commercially.
- Liability concentrates at the platform. Just like Napster, the entity that built the system and profits from it is the one in the crosshairs — not the anonymous corner of the internet the data came from.
- "Everyone did it" is not a defense. It wasn't for Napster, and it won't be here.
The labs feel this. That's why the frontier companies are now spending enormous sums on licensing deals with publishers, forums, and stock libraries. The market has already voted: clean, licensed data is worth paying for. The problem is that those deals are bilateral, slow, and only available to the handful of labs big enough to negotiate them.
The individual creator — the artist, the developer, the musician, the person with hard-won lived experience — has no seat at that table at all.
Music already solved this — a century ago
Here's the part that should give everyone hope. The music industry faced exactly this problem long before Napster, and it built a working answer.
A single songwriter cannot possibly track every radio station, bar, and venue that plays their song, let alone bill each one. So in 1914, ASCAP was formed, followed by BMI — performing-rights organizations that license entire catalogs collectively, collect the fees, and distribute royalties back to creators by usage. You don't negotiate with ten thousand songwriters to run a radio station; you get one blanket license, and the money flows to the right people automatically.
That model turned an unmanageable rights problem into a functioning, multi-billion-dollar economy that pays creators to this day.
AI training data needs the same institution. Individual creators can't negotiate with OpenAI, Anthropic, or Google. AI companies can't strike ten thousand individual deals. The thing that's missing is the collective in the middle.
That's Ché.
What ethical data actually looks like
"Ethical AI data" gets thrown around as a slogan. We mean something specific and verifiable. Ethical data has three properties, and you should refuse to call it ethical if it's missing any of them:
- Consent. The creator explicitly agreed to have their work included in licensable training datasets — not buried in a terms-of-service update, but as an affirmative, recorded act.
- Provenance. Every datapoint traces back to a specific, identifiable contributor through a record you can audit — not a marketing claim, an actual ledger.
- Compensation. When the data is used, the people who made it get paid, by usage.
Scraped data has none of these. A blanket "we trained on publicly available data" has none of these. This is the difference between ethically-sourced as a slogan and as an enforceable property — and enforceability is exactly what survives a courtroom.
How Ché builds it in
Ché is the collective licensing body for AI training data — the BMI/ASCAP for AI. People contribute the things that make us human: art, music, code, and lived experience. We capture it deliberately, attribute it permanently, license it to AI companies that want clean data, and pay the contributors royalties when their data is used. Here's how each ethical property is made real rather than promised:
Consent is an affirmative, recorded act. Before any contribution is anchored, the contributor makes an explicit ownership-and-rights attestation — "I created this, or I hold the rights to license it, and I grant Ché the right to include it in licensable datasets" — and an explicit "public & permanent" opt-in. Consent isn't assumed; it's signed.
Provenance is on-chain and tamper-proof. Every contribution is anchored to its author's address on-chain, with the content, a keccak256 hash, and a timestamp. The result is a permanent, public, auditable ledger proving every datapoint traces to a consenting human. An AI company licensing the Ché corpus doesn't have to trust that the data is clean — it can point to the immutable record and prove it.
Compensation flows by usage. When an AI company licenses the corpus, it pays a fee into the LicenseRegistry; a capped protocol cut routes to the treasury, and the remainder becomes a contributor royalty pool distributed on-chain by usage. We don't pretend a neural net lets us trace a single output token back to one person — instead we pay on a provable fact: was your contribution in the dataset snapshot that was licensed for this model? — weighted by tractable quality signals. Contributors claim their royalty trustlessly with a Merkle proof.
And critically, Ché is built to stay clean. A licensing body is only as defensible as its rights are. If a contributor uploads work they don't own and the true owner surfaces, Ché can flag the contribution, quarantine it from licensing the moment a claim is filed, escrow any royalties tied to it, and exclude it from every future dataset snapshot if the claim is upheld — clawing back funds and making the rightful owner whole. That dispute-and-takedown machinery is the difference between a platform that hopes its data is clean and one that can prove and defend it.
This is our answer to the Open Rights problem, and it's the structural opposite of Napster. Napster's architecture made it impossible to police infringement — that's literally why the court's filtering order killed it. Ché's architecture makes policing infringement a built-in primitive.
Why this is the winning position, not the cautious one
It's tempting to read "ethical data" as the slow, expensive, responsible-but-disadvantaged path. We think that's exactly backwards.
- For AI companies: a Ché license delivers (a) provenance you can prove in front of a judge, a regulator, or an enterprise customer's procurement team; (b) clean rights covering thousands of contributors in a single license; and (c) a defensible ethical story where creators are demonstrably paid. As copyright pressure mounts, that's not a nice-to-have — it's the difference between a model you can build a business on and one that's a liability waiting to be enforced.
- For creators: for the first time, the art, music, code, and experience that gets fed into AI is something you consent to, get credited for, and get paid for — collectively, with leverage you could never have alone.
- For the ecosystem: Ché keeps the two economies honest. $NEUROS is 100% fair-mined and rewards the people who make the data — zero premine, no founder allocation, because it's the contributors' token. $CORTEX funds and rewards the people who build the institution that licenses it.
Napster proved the demand for the medium and got destroyed by the rights problem. The companies that came after — the licensed ones — built the streaming economy that's worth tens of billions today. The value was always real. What changed was whether it was built on consent.
AI is at that same fork right now. Scraping was the Napster move: brilliant product, fatal foundation. Licensed, attributed, consented data is the streaming move — and it's where the durable companies of the next decade will be built.
We'd rather build the institution that pays creators than be the cautionary tale that didn't.
Visit Ché — che.cortexresearch.group