In the early 20th century, radio changed everything for musicians. Suddenly a song could be performed endlessly and on repeat across the country. From living rooms, diners, to hotel lobbies and parties — but the artist who wrote it would never see a cent. The technology moved faster than the infrastructure. Musicians had no way to know where their work was being used, and no mechanism to collect compensation when it was.
ASCAP was founded in 1914 to solve exactly this problem. BMI followed in 1939, deliberately opening its doors to blues, jazz, and country artists who were being shut out. What they built was the missing infrastructure: a neutral, creator-controlled layer that sat between the work and the world using it. Giving artists the ability to register their song, and every time it's performed publicly, they get paid. The system tracked usage, matched it to ownership, and distributed royalties at scale. It was unglamorous, technical, and foundational.
The creator economy in the AI era needs the same thing. We don't have it yet.
The Flood
Before AI became a cornerstone Gen Z lexicon, artificially generated content made up roughly 10% of what existed on the internet. After the launch of ChatGPT that number surged to 40% and has been climbing ever since, with some analysts projecting it could reach 90% by the end of 2026. That's not a gradual shift. It's a replacement event.
The effects are already visible. Feeds that once required human effort to populate are now saturated with what the internet has taken to calling “slop” — content that is technically coherent, emotionally empty, and completely inhuman yet optimized for algorithmic distribution. Trust is collapsing as a result. Only 26% of consumers now say they prefer AI-generated creator content, down from 60% in 2023. Audiences aren't just skeptical of AI content, they're actively moving away from it. Toward deliberately unpolished, human-made work as a signal of authenticity because perfection has become the mark of a machine.
The backlash is showing up across industries. Pearl Abyss, developer of the recently released Crimson Desert, issued a public apology after players discovered AI-generated 2D props in the game. A minor element in an otherwise massive production, but enough to trigger a genuine outcry with their audience. Nvidia's DLSS technology, which uses AI upscaling to insert AI-generated frames between rendered ones, drew sharp criticism from the gaming community including game designers and artists. On the other end, Take-Two — the parent company of Rockstar Games and the studio behind the most anticipated game in years, GTA 6 — made a point of publicly stating they are not using AI in development. Showing that authenticity and humanity are now a competitive advantage.
Social media platforms are caught in the middle. Nearly 1 in 10 of the fastest-growing YouTube channels in July 2025 consisted entirely of AI-generated content. Proving that the algorithm is rewarding volume over origin. The infrastructure to distinguish human-made from machine-made doesn't exist at the platform layer. And that gap is about to become an explosive problem.
What Exists Now — and Why It Isn't Enough
The closest thing we have to a provenance standard today is the Coalition for Content Provenance and Authenticity (C2PA). A Joint Development Foundation project between Adobe, Arm, Intel, Microsoft, Truepic, and more recently Google.
C2PA works via a claim generator. By utilizing a camera's firmware, Adobe Photoshop, or another certified tool it builds a structured record of facts about a file — such as who created it, what tools were used, when and where. Those “facts” are cryptographically signed and embedded in the file's metadata as a manifest and tagged with “Content Credentials,” creating a chain of custody from creation to publication.
Most major social media platforms including Instagram and TikTok strip file metadata on upload. Attaching an image to an email strips it. Taking a screenshot strips it. The credential exists inside the file, which means it only survives if every system in the distribution chain chooses to preserve it. And most actively remove it.
Structurally, while C2PA verifies that the manifest was signed it doesn't verify that the assertions contained within the manifest are true. Researchers at Hacker Factor demonstrated this by successfully creating a fully forged C2PA manifest attributed to a separate named individual using the c2patool. What C2PA does genuinely provide is a non-repudiation trail for participants who choose to sign honestly and operate under good faith. Which can be valuable infrastructure for journalists, news organizations, and hardware manufacturers. But it can't give users the one thing they actually want: confirmation that what they're viewing is real, and that it belongs to who it claims to belong to.
The rest of the toolkit unfortunately fares no better at the infrastructure level. robots.txt tells crawlers not to scrape your site and is almost universally ignored by the systems that matter most. Copyright registration gives you a legal asset, but only after a slow, expensive filing process, and it tells you nothing about when the work was created relative to what an AI company ingested. It is in no way built to handle the amount of content continually being produced. Watermarking gets bypassed when a model regenerates output. The University of Chicago's Glaze tool degrades style-mimicry in diffusion models, but the underlying provenance question — who made this, when, and under what license — remains entirely unanswered.
Finally, the platform licensing problem. When you post your work to Instagram, TikTok, or Facebook, you retain copyright on paper. What you have actually done is grant each platform an unconditional, irrevocable, royalty-free, perpetual, worldwide license to use, modify, adapt, reproduce, and distribute your work “in any format and on any platform, either now known or hereinafter invented.” That clause is standard across virtually every major platform's terms of service — written before generative AI existed. It now covers AI training. Ownership without control isn't ownership. It's a fiction that costs creators everything.
Each of these tools solves one surface of the problem. None of them constitute a structural layer.
What a Real Trust Layer Requires
So what would actually work?
Timestamp and certify at creation, before distribution.
The moment a work is finalized is when provenance needs to be established. Not after it's been uploaded to social media. Anything that happens after distribution is forensic work, not infrastructure. A standard that regulators are beginning to apply to AI companies — content, source, timestamp, opt-out status, license path, legal basis, immutability, audit trail — need apply at the creator level first, before the work ever enters the systems that will consume and eventually reproduce it.
Store it permanently and independently.
Platform-dependent provenance is not provenance. Platforms change their terms, get acquired, sunset features, and outright delete content. A trust layer has to live somewhere no single party controls, where nothing can be quietly removed. The record has to be durable in the same way a deed or a birth certificate is durable — not contingent on any company's continued participation or goodwill.
Make it machine-readable so AI systems and regulators can act on it.
The regulatory environment is starting to catch on. Starting in 2026, AI developers operating in the EU must check whether a data source carries a copyright reservation and exclude or license that content before using it in training. California now requires covered providers to embed machine-readable provenance data, with modifications that disable disclosure functions explicitly prohibited. The opt-out signal is becoming a legal obligation for AI companies to respect — but only if it exists in a form their ingestion pipelines can actually parse. A note on your portfolio page is not adequate.
The trust layer isn't a feature to be added to an existing platform. The creator economy has payment rails, distribution rails, and audience analytics. What it has never had is a verifiable, persistent, machine-readable record of who made what and under what terms. One that exists outside the platforms that profit from the content itself.
The law is important, but technology and markets move faster. What's needed are technical safeguards that operate at the data layer, not just legal frameworks that operate in courtrooms.
The music industry didn't solve its radio problem with lawsuits. It solved it by building a layer of record that existed before distribution, survived every platform that came after it, and gave creators something to point to when someone used their work without asking. ASCAP didn't start as a royalty system. It started as a durable link between a creator and their work that no distributor, label, or broadcaster could dissolve.
That's the model. And the creator economy in the AI era needs it more urgently than the music industry did in 1914, because the scale of ingestion happening right now makes early radio look like a trial run.
The infrastructure doesn't exist yet. Standards are being debated, and the platforms are making gestures toward authenticity while their terms of service quietly cover AI training. What's missing is the layer underneath all of it — permanent, independent, machine-readable — that establishes the record before anyone else gets to decide what your work is worth or who it belongs to.
Read more about how AI training data pipelines actually work and why why Stelais chose Arweave for permanent storage.