The AI industry stopped arguing and started paying. Getty proved permission has a price. Nobody has yet proved that origin does.
The photograph at the top of this page is doing more economic work than almost anyone in the value chain will ever be paid for.
Two young women, a railing, a phone, a laugh. It is a stock image. It sits in a library alongside hundreds of millions of others. Somewhere it has been licensed to a bank, a telecom, a health insurer, a government campaign — each of which needed exactly this: the visual signal of young, Black, urban, connected, joyful. That signal is not incidental to the sale. It is the sale. It is why the image was commissioned, why it was keyworded, why it ranks, and why it costs what it costs.
Now the same image goes to work again. It is retrieved inside an AI answer engine. It becomes a reference for what "friendship" or "Gen Z" or "mobile banking" looks like. It shapes a model's sense of a people. And in 2026, for the first time at scale, somebody actually got paid for that second life.
Just not the two women. Not their neighborhood. Not the culture that made the aesthetic legible enough to be worth licensing in the first place.
This note is about the gap between those two facts — and about the eighteen-month window we have to close it before it hardens into standard practice for every museum, archive, and public collection on earth.
The market has learned to pay for permission. It has not yet learned to pay for origin. MADE CX exists to make the second as contractually routine as the first.
Two deals, eight months apart, converted a legal argument into a business model. What changed was not the law. What changed was the tape.
For three years the content industry fought AI on a single question: is training on our work theft? The answer, in courtroom after courtroom, has been maddeningly partial. Then the argument moved venues — out of the docket and onto the ticker.
On 31 October 2025, Getty Images and Perplexity signed a multi-year global image partnership: Getty content, delivered through Getty's API, surfaced inside an AI answer engine, with a commitment to display "image credit with link to source."2 On 20 June 2026, Getty announced a multi-year display partnership with OpenAI, putting its licensed library into ChatGPT's search and discovery results.1
The market's verdict was immediate and unambiguous. Getty stock rose roughly 145% on the OpenAI news — reported by some outlets as more than a doubling in early trading.3 A company that had spent three years being described as an AI casualty was repriced overnight as an AI supplier.
The courts did not deliver "license, don't ingest." The balance sheet did.
Section 01Read the sequence carefully, because the order matters. Getty's litigation largely failed. Getty's licensing spectacularly succeeded. Rights holders everywhere drew the obvious conclusion: the durable protection is not a lawsuit, it is a contract — and the contract is worth writing because someone will now pay for it.
That is a genuine victory. I want to be clear that I think it is a good thing, and that Getty earned it by refusing to fold when folding was the consensus advice. But a norm is only as good as its clauses. And when you read these particular contracts for what they don't say, a second market failure comes into focus — one that the first victory has now made much harder to see.
Both deals answer the question "may we?" Neither answers the question "whose?" That distinction is the entire content of this paper.
A license is a transaction between whoever holds the file and whoever wants the file. It is a clean, ancient, well-understood instrument. Its one structural blindness is that it asks nothing about how the holder came to hold.
When Getty licenses an image of two young women in a Black neighborhood to an AI platform, four distinct forms of value move at once:
One: the photographer's craft — framing, light, moment. This is copyright. It is registered, it is titled, and it is paid, at a contributor rate.
Two: the subjects' likeness — their faces, their bodies, their presence. This is a release. It was signed once, for a fee, in perpetuity, before anyone in the room knew what "training data" meant.
Three: Getty's aggregation, curation, keywording, indemnity, and distribution. This is real work and it is properly compensated. It is most of what the AI platform is actually buying convenience from.
Four: the culture itself — the aesthetic vocabulary that makes the image legible and commercially desirable. The hair. The posture. The color. The specific register of joy. The reason a brand chose this image over ten thousand technically identical ones. This is the input with the highest marginal commercial value in the frame, and it is the only one with no title, no registry, no rate card, and no counterparty.
Layers one through three are property. Layer four is treated as weather — an ambient condition, free to all, owned by none. So it is priced at zero, and everyone downstream books the difference as margin.
You cannot pay what you cannot identify. And nothing in the current deal architecture identifies a community of origin — so nothing in it can pay one.
This is not a failure of goodwill on anyone's part. It is a failure of instrumentation. Getty's contracts are excellent at what they were built to describe. They were simply never built to describe this.
There is a second omission worth naming, because it will matter enormously in the museum context. The OpenAI arrangement is publicly framed as a display partnership. Whether Getty content trains OpenAI models is, on the public record, unstated.1 Neither release discloses financial terms. Neither release describes how — or whether — the roughly 600,000 content creators in Getty's contributor network share in this revenue.42
Display, retrieval, fine-tuning, pre-training, and synthetic derivation are five economically distinct acts with five different half-lives. Collapsing them into one word — "partnership" — is how value quietly migrates. A standard that does not force them apart is not a standard. It is a press release.
Strip Getty to its balance sheet and you find a holding company for other people's culture. That is not an accusation. It is a description — and it is the clearest proof we have that cultural property is already a priced asset class.
By our assessment, roughly 76% of Getty's enterprise value derives from selling, holding, and commercializing cultural property — imagery of people, places, movements, moments, and aesthetic traditions that the company did not originate.5 The public financials support the shape of that claim even where they cannot confirm the precise figure.
In its thirtieth year Getty booked record revenue of $981.3 million — Creative $556.9M, Editorial $369.6M, Other $54.8M — with adjusted EBITDA of $320.9M at a 32.7% margin. The library runs to 609 million images and 36 million videos, sourced from over 600,000 content creators and more than 360 content partners.4
Look at the composition. Creative and Editorial together are 94.4% of revenue. Both are, in substance, the licensing of depicted human life. Editorial is the record of what people did. Creative is the manufactured aesthetic of who people are. Neither is a technology product. Both are cultural property, warehoused and rented.
The 76% figure is our estimate of how much of the enterprise value — not the revenue line — sits in that warehouse rather than in the platform, the tooling, or the contracts around it. Reasonable people will argue the number. Nobody serious argues the direction. And the OpenAI repricing settled the matter empirically: the market added billions in value on the announcement of a deal whose entire subject matter is access to depicted culture.3
Getty did not create the culture in 609 million photographs. It created the index. The market pays for the index, and the index keeps the money.
Section 03I want to be precise about what I am and am not saying. I am not saying Getty is a villain. Getty is a well-run intermediary doing exactly what an intermediary is supposed to do inside the rules as written. It has litigated for rights holders at real cost. It has built the discovery, clearance, and indemnity infrastructure that makes the market function at all. Under CPRS, an intermediary like Getty is not eliminated — it is registered, and its commission becomes transparent rather than residual.
What I am saying is that the rules as written contain a hole the size of a civilization, and that we are about to pour the world's museum collections straight into it.
Read the Getty–OpenAI and Getty–Perplexity announcements as a lawyer reads them: not for what is promised, but for what is conspicuously absent.
Perplexity's commitment is real and worth acknowledging: image credit with a link to source, so users learn how licensed imagery is legally used. Getty's Nick Unsworth framed the deal as one that "acknowledges the importance of properly attributed content." Perplexity's Jessica Chan called attribution and accuracy "fundamental to how people should understand the world in an age of AI."2
Attribution to whom, though? To Getty. The credit line resolves to the intermediary. In the entire public record of both agreements there is no mechanism that names, credits, compensates, or gives standing to the people and communities the imagery depicts and derives from.
Here is the audit, clause by clause.
| Governance function | Present in the Getty AI deals? | Consequence |
|---|---|---|
| Permission to use | ✓ YES — the deals exist precisely to establish this | The genuine achievement. Ingestion-by-default is dead. |
| Attribution | ✓ PARTIAL — credit-with-link, resolving to Getty | Credit flows to the index, not the origin. |
| License-class separation (display / retrieval / training / synthetic) | ✗ NO — training status undisclosed; acts bundled as "partnership" | The most valuable right may transfer without ever being priced. |
| Community-of-origin identification | ✗ NO — no registry field exists for it | No counterparty can be paid because none is named. |
| Royalty waterfall to originators | ✗ NO — terms undisclosed; contributor share unstated | 600,000+ creators cannot verify their own position. |
| Community reinvestment floor | ✗ NO | Value exits the culture permanently on first sale. |
| Asset-level usage reporting | ✗ NO — no public telemetry commitment | Nothing to audit means nothing to enforce. |
| Audit rights | ✗ NO | Trust substitutes for verification. |
| Term limits & renegotiation triggers | ✗ NO — "multi-year," undefined | Terms set in 2026 govern model generations nobody has imagined. |
| Community standing to enforce | ✗ NO | The depicted have no seat, no voice, no remedy. |
One row filled. Nine rows empty. That ratio is the market failure, rendered as a contract.
And every one of those nine gaps is fillable. None requires new legislation. None requires a court to rule our way. Each is a clause that a sufficiently determined counterparty can insist upon at the negotiating table — which is exactly why the identity of the next counterparty matters so much.
Frame it as ethics and you get a statement of values. Frame it as missing infrastructure and you get a system that pays.
Every functioning market rests on a registry. Land has title deeds. Companies have share registers. Music has ISRC and publishing splits. Shipping has bills of lading. Accounting has GAAP. None of these were inevitable; each was built, and each was resisted, and each converted a chaotic domain into a financeable one.
Culture has nothing. There is no title, no chain of custody, no standardized disclosure, no royalty accounting, and therefore no enforcement. The result is not that culture goes unmonetized — it is monetized relentlessly. The result is that the monetization is uncontested, because contest requires standing, and standing requires a record.
This is the same structure Hernando de Soto described in the informal economies of the global south: enormous stocks of real, valuable, actively used assets that cannot function as capital because they are not registered. Dead capital. The culture economy is the largest pool of dead capital in the world, and the AI transition is the moment it either gets titled or gets permanently absorbed into somebody else's balance sheet.
Current value flow — unstandardized
CPRS value flow — standardized commerce
In markets, what is not measurable is rarely enforceable. CPRS makes cultural property measurable — and therefore governable.
Registration does not moralize the transaction. It instruments it. Once an asset carries an identifier, a provenance record, a license class, and a waterfall, the payment is no longer a favor anyone is asking for. It is a term of the deal, auditable like every other term.
Getty licensed its own commissioned library. Museums and archives are about to license something categorically different: collections of other people's culture, much of it acquired under conditions no modern contract would permit.
In February 2026 the UK confirmed that twelve national institutions will take part in a government-backed AI marketplace pilot — the Creative Content Exchange — with an operational platform expected in summer 2026.6 The participants are not marginal: Historic England, Imperial War Museums, the National Library of Scotland, the Natural History Museum, The National Archives, the National Portrait Gallery, Oxford University Museums, the Royal Armouries, Royal Botanic Gardens Kew, Royal Museums Greenwich, the Science Museum Group, and the V&A. The scheme sits inside the R&D Missions Accelerator Programme announced with a £380 million package.
The stated purpose is to let AI developers access digitized cultural assets "while respecting the rights of creators and copyright owners." Revenue-sharing arrangements are not specified in the public material.6
Read that phrase again: creators and copyright owners. For a museum collection, the copyright owner is very often the institution. The creator is very often dead. And the community whose material culture, ancestors, sacred objects, botanical knowledge, portraits, and colonial-era photographic record constitute the actual asset is — again — not a category the sentence contains.
A national archive licensing its holdings to a foundation model is not selling its own culture. It is selling its custody of everyone else's.
Section 06This is why the museum window is the decisive one, and why I am writing this note in August 2026 rather than in 2028.
First, the asset is qualitatively different from Getty's. Getty's library was commissioned, released, and contracted. A national collection contains material gathered across two centuries of expedition, empire, excavation, purchase, bequest, and seizure. Provenance in a museum is not a metadata field; it is the institution's most contested public question. Licensing that material into a training corpus without a provenance clause does not merely repeat the original acquisition — it makes it permanent, machine-readable, and infinitely reproducible.
Second, these institutions are governed by mandates that already point our way. Getty answers to shareholders and is right to. A national museum answers to a public trust, a repatriation policy, a community engagement charter, and in many cases an explicit decolonization commitment. The clauses I am about to propose are not a burden foisted on these institutions. They are the contractual expression of duties they have already publicly accepted and currently have no mechanism to honor in a data deal.
Third — and this is the whole strategic argument — the first template becomes every template. Standards do not propagate by merit. They propagate by precedent and procurement. Whatever clause set the Creative Content Exchange pilot settles on will be the base document for the Smithsonian, for European nationals, for university special collections, for African and Caribbean institutions negotiating with the same handful of counterparties from a far weaker position. If the first cultural-data license in the world contains no origin field, no cultural-data license will contain one for a decade.
I do not think we have eighteen months. I think we have this fiscal year.
We are not asking institutions to be more culturally sensitive. We are handing them the clause set that makes the sensitivity they have already committed to legally operable, auditable, and priced.
CPRS was built for exactly this ecosystem — creators, estates, museums and archives, brands and platforms — and the adoption path for institutions was designed before this pilot was announced. The infrastructure is not hypothetical. The counterparty is.
This is the operative content of this note. Below is the MADE CX governance rider for a cultural-data license — designed to be appended to a deal that already looks like Getty–OpenAI, without renegotiating the commercial core.
Deliberately, none of these clauses requires new law, a court victory, or the counterparty's agreement that anything was ever taken. Each is a term an institution with leverage can put on the table in the ordinary course — and in the current market, an institution holding a unique national collection has more leverage than it has had in a century.
No asset enters the licensed corpus without a registry record naming the originator, the estate or successor where applicable, and the community of origin. Unattributable assets are flagged, not silently included.
Display, retrieval/RAG, fine-tuning, pre-training, and synthetic derivation are five distinct grants. Each is separately consented, separately priced, and separately reported. Silence grants nothing.
The record includes chain of custody and the circumstances of acquisition. Contested, restricted, sacred, and repatriation-flagged holdings are marked and excluded by default from any training-class grant.
Credit must survive embedding. Machine-readable attribution travels with the asset into the corpus and surfaces at output level — naming originator and community of origin, not only the licensing institution.
Defined territory, term, channels, derivative rights, and exclusivity. No perpetuity. No assignment on change of control. Automatic renegotiation on a material change in model generation or deployment context.
A defined, disclosed distribution across originator, estate, custodial institution, servicer, and community fund — published in the deal, not buried in a side letter.
Quarterly usage telemetry at the individual asset level, by license class. Aggregate reporting is not compliance. Reporting obligations survive termination for the life of any model trained under the grant.
An independent audit right exercisable annually, with costs shifting to the licensee on a material discrepancy. Without cost-shift, an audit right is decorative.
The community of origin is a named party to the instrument with independent standing to enforce — not a third-party beneficiary of institutional goodwill. Reinvestment vehicles are governed by local stakeholders.
The license conveys use, never title. No clause may be construed to transfer, exhaust, or extinguish cultural property rights. Unlicensed use is a defined compliance breach with a proportional, documented remedy ladder.
Note what the rider does not do. It does not set the price. It does not tell the institution who to deal with. It does not require the AI developer to concede any historical claim. It does not slow the transaction — clauses 01 through 04 are metadata work the institution's registrars are already doing, now given commercial consequence.
What it does is make the deal legible. And a legible deal is a financeable one: a corpus with registered title, disclosed provenance, separated license classes, and audited cash flow is worth materially more to a serious counterparty than an undocumented one, because it carries less indemnity risk. Governance is not a discount. It is a premium.
A direct comparison of the market-standard AI content license as demonstrated in 2025–26, against the same transaction executed under a MADE CX governance rider.
| Deal term | Market standard (2026) | Under MADE CX governance |
|---|---|---|
| Subject matter | "Content" — an undifferentiated library | Registered assets, each with a BCID title record |
| Grant | One bundled "multi-year partnership" | Five separately priced license classes; silence grants nothing |
| Provenance | Not a contract term | Disclosed; contested holdings excluded by default |
| Attribution | Credit-with-link, resolving to the intermediary | Machine-readable, output-level, resolving to origin |
| Financial terms | Undisclosed | Waterfall published as a term of the deal |
| Originator share | Unstated; contributors cannot verify position | Target 80% to originator side, on the ledger |
| Community share | None | 4% mandatory reinvestment floor, locally governed |
| Reporting | None disclosed | Quarterly, asset-level, by license class |
| Audit | None disclosed | Annual, independent, with cost-shift on discrepancy |
| Term | "Multi-year," undefined | Defined term; sunset and renegotiation triggers |
| Enforcement | Litigation, after the fact, by the intermediary | Compliance breach with a defined remedy ladder |
| Community position | Absent | Named party with independent standing |
| Title | Ambiguous at the margins | Never conveyed; non-extinguishment is express |
| Downstream effect | A press release | An auditable cash-flow history — securitization-ready |
The right-hand column is not aspirational. Every row is a clause that exists in commercial practice somewhere — in music publishing, in pharmaceutical benefit-sharing, in extractive-industry community agreements, in indigenous data sovereignty protocols. Nothing here is novel. The novelty is only in applying it to culture, which the market has never treated as property because we never built the register.
We are not asking for a new right. We are creating the paperwork that every other asset class received a century ago.
Section 08CPRS was designed for the full ecosystem of cultural commerce — creators, estates, museums and archives, brands and platforms. This is what adoption looks like for a collecting institution entering an AI licensing arrangement.
Steps 01 through 03 are, for most national institutions, work already underway under a different name — collections documentation, provenance research, and access policy. CPRS does not add that burden. It gives that burden a revenue line.
A counterparty that executes under this rider earns a compliance signal: verified attribution, verified provenance, verified reinvestment.
For an AI developer facing a fragmented global copyright landscape and unquantifiable indemnity exposure, a registered, audited, provenance-clean corpus is not a concession. It is the cheapest risk transfer available.
I would rather publish the strongest versions of the counter-case than pretend it doesn't exist.
This is the serious objection, and it deserves a serious answer rather than a dismissal. Cultural boundaries are porous, contested, and genuinely hard. Any registry will make judgment calls that somebody reasonably disputes.
But notice that every registry in history faced exactly this problem and was built anyway. Land title had to resolve overlapping customary claims. Music publishing had to resolve co-writes, samples, interpolations, and traditional arrangements — and now settles billions annually across splits that are frequently disputed and routinely amended. The existence of hard cases is an argument for a dispute mechanism, not for the absence of a register.
CPRS handles this with a verification layer and a documented conflicts process. Contested claims are surfaced and adjudicated rather than silently resolved in favor of whoever currently holds the file — which is what "unworkable" actually means in practice today: the default resolution is always zero.
This assumes the buyer's only variable is price. In a market where the same buyer is simultaneously defending copyright litigation across multiple jurisdictions and cannot quantify its own corpus exposure, provenance-clean supply commands a premium, not a discount.
Getty's own repricing is the proof. The market did not reward Getty for having cheap content. It rewarded Getty for having defensible, indemnifiable, licensable content at a moment when the alternative carried unbounded legal tail risk. A registered corpus is that argument, extended.
And there is a floor effect: the twelve institutions in the UK pilot are not interchangeable suppliers. There is no second National Portrait Gallery. Uniqueness is leverage, and leverage unexercised is leverage surrendered.
Correct — and that is precisely the point. Copyright covers the fixed expression: the photograph, not the aesthetic; the recording, not the tradition; the catalogue record, not the object's meaning to the people it came from. Those excluded layers are where the commercial value increasingly sits.
We are not asking a court to invent a right. We are proposing a contractual standard between consenting parties — the same mechanism by which every industry standard, from GAAP to ISO to SWIFT, became binding long before any statute mentioned it. CPRS is enforceable because parties agree to it, and parties agree to it because it reduces their risk.
The UK judgment in Getty v Stability makes this the only realistic path. The court found model weights are not stored copies, and Getty's training claim failed on jurisdiction. If the strongest-resourced rights holder in the sector could not win on copyright, waiting for copyright is not a strategy.7
It is not, and the distinction is load-bearing. Reparations are backward-looking, involuntary, and adjudicated. CPRS is forward-looking, contractual, and prospective: it governs the next license, not the last two hundred years. It makes no historical finding and demands no admission.
Reginald Lewis did not argue with Wall Street about whether Black people deserved access to capital. He arrived with deal structure. That is the posture here. We are not asking brands, platforms, or institutions to be more culturally sensitive. We are installing the infrastructure that makes unlicensed use of cultural property a compliance issue — with the same weight copyright already gives to music, film, and software.
If a governance standard that pays originators and communities happens to correct a historical imbalance, that is a welcome consequence. It is not the mechanism, and it is not the pitch.
Here is what I believe, stated plainly.
The fight over whether AI companies should pay for content is over, and the content industry won it. That victory is real and I do not want to diminish it — for three years the people arguing for permission were told they were standing in front of a train, and the train stopped and bought a ticket.
But the ticket has one name on it, and it is the name of the intermediary. Getty won. The 600,000 contributors behind Getty's library have, on the public record, no disclosed share of what Getty won. The people in the photographs have nothing. The communities whose aesthetic makes those photographs worth licensing have less than nothing — they have a norm now hardening around them that says their contribution is the one input in the chain that arrives free.
That is the market failure. Not malice. Not extraction by design. Just the ordinary, grinding consequence of an asset class that was never registered, being sold by everyone who touched it except the people who made it.
Trillions in economic value will move through cultural data licensing over the next decade. Under the current template, essentially all of it accrues to intermediaries and platforms — because they are the only parties the contract can see.
And the template is being written this summer, in a government pilot, by twelve institutions holding collections assembled from the entire world.
Whoever writes the first cultural-data license writes all of them. That document is on somebody's desk right now, and it has a blank where the community should be.
Founder's Note No. 07So the ask is narrow and it is practical.
To museums, archives, and national collections: before you sign, append the rider. Ten clauses. None of them contests your title, delays your timeline, or costs you your counterparty. All of them convert the ethical commitments already in your public charter into terms that survive contact with a procurement department.
To AI developers: a registered, provenance-clean, audited corpus is the cheapest indemnity you will ever buy. You are already paying for permission. Pay a little more and buy certainty.
To creators, estates, and communities: register. A claim without a record is a feeling. A claim with a record is an asset.
License, don't ingest — we won that. Now: license to the origin, not just to the index. That one is still open, and it will not be open long.
Yours in the interest of the race,
Where $CULTURE is titled, priced, and governed as financial-grade property. Subscribe to get each Founder's Note sent to your inbox.
Verified public figures are cited below. Where a figure is our own estimate or computation rather than a reported number, it is labeled as such.
MADE CX publishes editorial research and commentary on cultural property and commerce. Nothing on this site is an offer to sell or a solicitation of an offer to buy any security or interest, and nothing here is investment, legal, accounting, or tax advice.
Designations shown with a dollar sign — for example $CULTURE — are editorial identifiers used to name the subject of an analysis. They are not securities, digital assets, tokens, funds, or tradeable instruments of any kind, and no such instrument is offered or available.
Names, marks, and images of individuals, estates, and organizations appear for purposes of news reporting, commentary, and analysis. Their appearance does not imply affiliation with, sponsorship by, or endorsement of MADE CX.
© 2026 MADE CX. MADE CX, CPRS, and Culture Exchange Standard are trademarks of MADE CX. All other marks belong to their respective owners. Corrections and rights inquiries: hi@madecx.info.