Where $CULTURE fits in the AI vertical.

Culture goes in. Capability comes out. Value lands everywhere except the source.

Every frontier model is a compression of human expression. The language it speaks, the music it can generate, the images it can render and the taste that makes its output commercially usable all arrive from somewhere. A large and disproportionate share of that somewhere is culture made by communities that have never held title to what they made.

The AI economy has already proven that cultural inputs carry a price. Between 2023 and 2026, AI developers signed licensing agreements with news publishers, stock libraries, social platforms and major record labels, and one developer agreed to pay authors roughly $1.5 billion to settle a class action over pirated books.¹⁴ Those payments did not go to culture in general. They went to parties that could prove ownership.

This paper makes four arguments. First, the AI value stack captures value at the compute, model and distribution layers while treating the cultural layer as free feedstock. Second, the parties positioned to benefit from that arrangement are structural, not villainous, and they will keep benefiting until title exists. Third, culture is mispriced because it lacks three things every other asset class has: a registry, a valuation standard and market infrastructure. Fourth, the Cultural Property Rights Standard (CPRS) supplies all three, and its revenue design returns value to the communities that produced it.

The commercial ecosystem that generates and captures value.

To see where $CULTURE fits, start with how the AI industry is built. It is a stack of five layers, and money moves up and down it in very different amounts.

1.1Where value is created and where it is captured

Capital in the AI economy concentrates where scarcity is legible. Chips are scarce and metered by the unit. Cloud capacity is scarce and metered by the hour. Model access is metered by the token. Each of those layers has a unit, a meter and an invoice.

The cultural layer has none of the three. It is the reason a model can write in a recognizable voice, produce a beat with a regional swing, or render a photograph that reads as authentic to a place. Yet in the accounts of the companies that build on it, culture appears as a scraping cost or a legal reserve, not as a cost of goods. Value created at the bottom of the stack is booked at the top.

1.2Culture as unpriced feedstock

Training data is commonly treated as an externality: something that exists in the world and can be collected. Three features of culture make that treatment especially costly for its originators.

Culture is the taste layer. Raw capability is not enough to make generative output sellable. Output has to feel current, specific and human. That feeling is learned from style, vernacular and aesthetic lineage, which are cultural properties.

Culture travels as style, not as copies. Copyright protects specific expression. A model trained on a tradition can produce new work "in the style of" that tradition without reproducing any single protected work. The commercial value moves; the legal hook does not.

Culture is often collective. A genre, a dialect, a design vocabulary or a dance form has many authors across generations. Individual-copyright frameworks were not built to recognize that kind of ownership, so it defaults to no one.

A model trained on a tradition can sell that tradition back to the world without copying a single protected work.

1.3The licensing market proves the asset exists

The strongest evidence that culture is an asset is that AI companies already pay for parts of it. Reported deals include Reddit's agreements with Google and OpenAI, News Corp's multi-year agreement with OpenAI, the New York Times' agreement with Amazon, and stock-image licensing by Shutterstock to multiple model developers.⁴² In music, Universal Music Group settled its suit against Udio in October 2025 and agreed to build a licensed platform with it, and Warner Music Group reached settlements and license agreements with both Udio and Suno shortly after.³⁵

Litigation tells the same story from the other side. The 2025 ruling in Bartz v. Anthropic found that training on lawfully acquired books may qualify as fair use, while letting claims over pirated copies proceed; the parties then settled for at least $1.5 billion, about $3,000 per work.¹ The New York Times suit against OpenAI and Microsoft, filed in December 2023, and Getty Images' 2023 suit against Stability AI show rights holders using title as leverage.²

Who benefits from the market failure.

This section describes positions, not motives. Most of the companies below are doing what any firm does in a market without property lines: using the inputs available at the lowest cost the law allows. The point is structural. As long as cultural title does not exist, each of these positions earns more by leaving it that way.

2.1Compute and infrastructure

Chipmakers and hyperscale cloud providers sell the capacity every training run consumes. Their revenue rises with the volume of data processed, regardless of where the data came from or whether anyone was paid for it. They carry little legal exposure for training inputs and capture margin on every cycle.

2.2Foundation model developers

Frontier labs and open-weight developers convert cultural inputs into general capability. Cultural fluency, the ability to sound right and look right across communities, is a headline product feature. These developers are also the parties most exposed to litigation, which is why they have become the main buyers in the licensing market.

2.3Platforms and aggregators

Social, streaming and user-generated platforms host enormous volumes of culture created by their users. Their terms of service typically grant the platform broad license rights, which lets the platform resell or license that corpus to model developers. Reddit's reported AI agreements are the clearest public example of user culture becoming platform data inventory.⁴

2.4Generative application companies

Music, image, video and voice generators sell output that is valued for its resemblance to existing styles and sounds. Their product is, in effect, a stylistic replica on demand. The music settlements of late 2025 show that this layer can be brought into license once a rights holder with standing appears.³

2.5Legacy rights holders

Major labels, large publishers and stock libraries are beneficiaries of a different kind. They hold registered title, so they are the parties AI developers negotiate with. Some of what they own originated in communities that were never paid fairly at the time. Their deals are useful evidence: title, not authorship, is what the market rewards.

2.6Dataset assemblers and brokers

Scraped corpora, curated datasets and the intermediaries who package them sit between the open web and the model developer. They add value through cleaning and labeling, but the underlying material is mostly culture collected without a license.

Table 1 · MADE CX assessment of structural position. "Nothing" describes value flowing without CPRS attribution, not a finding that any party underpaid.

Why culture is mispriced, and why AI makes it urgent.

3.1Three missing pieces

Every recognized asset class rests on the same three pieces of infrastructure. Real estate has the deed registry, the appraisal and the closing. Equities have the share register, the analyst rating and the exchange. Music publishing has the copyright office, the rate court and the collecting society.

Culture lacks all three. There is no registry that records who originated a style, a sound or a vocabulary. There is no valuation standard that lets one cultural asset be compared to another. And there is no market infrastructure, the layer that defines any asset category, for licensing, clearing payments and enforcing authorized use. Without those, a buyer cannot find a seller, a price cannot form, and the default price is zero.

3.2The extraction pattern, historically

The current moment is not new. The BLK:INV Terminal, MADE CX's database of approximately 1.99 million historical Black innovation patent records, benchmarks $293 trillion in Lost Commercial Value platform data · modeled. The pattern in those records repeats across music, fashion, language and design: commercial value is realized downstream, attribution is lost upstream, and the originating community holds no equity in the result.

AI runs that same pattern faster and at global scale. A style that once took a decade to move from a neighborhood to a national campaign can now move from a training set to millions of generated outputs in a single product release.

AI is the old extraction pattern running at machine speed.

3.3What AI changes economically

The cost of imitation approaches zero. When a model can produce a convincing piece of work in a given style for a fraction of a cent, the market price of commissioning the originator falls with it.

Imitation substitutes for the original. Earlier copying competed with originators. Generative output can replace them in the same transaction: the stock photo, the jingle, the voice-over, the illustration.

Value shifts from works to corpora. AI buyers license libraries, styles and voices, not single songs or images. The unit of value is moving toward the kind of collective asset culture already is.

Leverage decays after training. A rights holder negotiating before a model is trained can withhold data. After training, the value is already embedded. Time is the variable that most affects what Culture Makers can capture.

3.4Capitalization, not only cash flow

Licensing royalties are the visible lever, and the smaller one. MADE CX modeling of its 2050 wealth-gap thesis attributes roughly 75 percent of the outcome to capitalization: recognizing culture as a balance-sheet asset that can be rated, financed, collateralized and indexed MADE CX model. A royalty pays once per use. A recognized asset can be borrowed against, held in an estate and benchmarked in an index. The $CULTURE Index is designed for that second function.

Across the full commercial use of Black cultural production, MADE CX estimates approximately $15 trillion in value generated without priced attribution MADE CX estimate. That figure describes unpriced value, not a claim that any specific party owes it.

3.5Counterarguments, addressed

Fair use. U.S. courts have signaled that training on lawfully acquired material can be transformative.¹ CPRS does not depend on winning that argument. It works on the commercial side: licensed, documented provenance is worth more to AI buyers facing regulatory and reputational exposure than an unlicensed corpus, whatever the fair use outcome.

Attribution is technically hard. Tracing a single output to a single input is unreliable. CPRS licenses at the corpus and style level, the same level at which current AI deals are written, rather than claiming per-output attribution.

Collective works have no owner. Collective rights management already exists for music performance. CPRS applies the same model, with custodians and communities able to register lineage-based claims recorded under the Heritage & Lineage dimension.

Policy is fragmented. The U.S., the EU and African jurisdictions are moving at different speeds. A private, auditable standard that operates under contract does not have to wait for harmonization, and it can feed into legislation where it is ready, as the Maryland CPRS effort shows.

A clearing system for $CULTURE.

The Cultural Property Rights Standard is MADE CX's open standard for how cultural property is recorded, evaluated, licensed and enforced as an asset class. It does for culture what credit ratings did for debt: it gives a buyer a consistent, auditable way to compare one asset with another before money moves.

4.1The rating

Every CPRS record is scored on five dimensions, each with evidence that moves it up or down.

Two constants matter most for this paper. Tech and AI Data carries the highest sector multiplier in the standard, 4.5, which reflects how intensively that sector consumes cultural inputs. And the AI extension applies a further 2.0 to 5.0 times factor to a record's modeled band when the asset is licensed for training or synthetic use. Both are modeled bands, never prices, and every published figure carries its pass type and confidence.

4.2The infrastructure stack

BCID (Blackchain Creative ID) establishes identity and provenance: who made the work and who holds it. CPRS rates it. Cultural Use Licenses set the terms under which AI developers and others may use it. The BLK:INV Terminal and the $CULTURE Index supply market data and benchmarks. Together they give culture the registry, valuation standard and market infrastructure described in Section 3.1.

4.3How it plugs into the AI vertical

Corpus-level training licenses. A collection, archive or estate registers once and offers a training license tiered by its CPRS record. This matches how current AI deals are structured.

Style and voice licenses. Generative applications can license a registered style, voice or likeness for a defined use, with the Likeness & Identity dimension setting the base.

Usage reporting. Licenses carry reporting duties tied to the registered record, so use becomes data that feeds back into the Commercial Usage dimension.

Enforceable opt-out. A registry entry turns "do not train on my work" from a request into a documented position with a named record, a published license and a counterparty.

4.4The regenerative mechanism

Licensing alone can repeat the old pattern with a receipt attached. CPRS is designed so that value recirculates. Every licensed transaction is split three ways.

The CX Foundation, the nonprofit arm of the system, reinvests its share into research, policy, education and community return programs aimed at improving conditions in Black communities. That reinvestment produces more registrations and better data, which raises the quality of every rating in the system.

1. STEP 1
   Register

   BCID establishes title and provenance.

2. STEP 2
   Rate

   CPRS sets a comparable, auditable record.

3. STEP 3
   License

   Cultural Use Licenses define authorized AI use.

4. STEP 4
   Return

   80/16/4 split pays the originator first.

5. STEP 5
   Reinvest

   CX Foundation funds research, policy and education.

6. STEP 6
   Grow

   More records sharpen the $CULTURE Index.

4.5Policy alignment

CPRS is built to run under private contract today and to inform public law where legislators are ready. In Maryland, MADE CX is working toward state legislation on the Baltimore Standard, which would make CPRS a state-level template for recognizing cultural property. Internationally, Africa's developing collective rights and cross-border licensing work under the AfCFTA framework addresses the same problem at continental scale, and MADE CX treats that legal layer as a counterparty to work with.

For Culture Makers: become a counterparty before the window closes.

Training cycles, licensing precedents and policy windows are open at the same time, and they will not stay open. Each new model release embeds more of the culture already available to it. Each new licensing deal sets a precedent for who gets paid. The decisions Culture Makers make in the next few cycles will decide whether they enter this market as owners or remain in it as raw material.

1. Register

   Get a BCID and title your work, archive, estate or collective tradition. Register any AI sub-properties, such as voice or style, before digitization or release.

2. Rate

   Get a CPRS record so you know what you hold and can show a buyer a comparable, auditable valuation band.

3. License on your terms

   Set Cultural Use License terms for training, style and voice use, including terms you will not accept.

4. Organize

   Register collectively. A catalog of many Culture Makers has far more leverage at the table than a single file, as every collecting society has shown.

The ecosystem has a role for everyone who cares about culture as property. Culture Makers originate and register. Culture Stewards hold archives and estates in custody. Culture Enthusiasts build the demand and the record of use. Buyers, including AI developers, get licensed provenance they can defend to regulators, investors and their own users.

The Culture Exchange is open.