How can proprietary data be liquidated?

Authors: Mohamed Tah¹

Affiliations: Voidback, Inc²

Keywords: Data, Commodities, AI

Abstract

Proprietary data held by a government, organization, or person is in high demand from AI labs, yet it remains illiquid: legal liability and consent requirements make it hard to exchange, and copying breaks scarcity. This paper asks what it would take for private data to trade as a commodity. Drawing on how oil and wheat became liquid, it argues for three verifiable pillars: provenance, compliance, and quality. It maps existing mechanisms to each pillar and sketches a regulated exchange in which access, not ownership, is the product. It also covers the limits:

Fair compensation is unsolved, and re-identification risk can be reduced but never eliminated. It concludes that a narrow market, built one well-defined category at a time, is feasible.

1. Introduction

Proprietary data is privately held data belonging to an entity; that entity could be a government, an organization, or a person.

Proprietary data is in high demand (Micro1 pays millions, along with other labs); Meta attempted to record its engineers' activities to train the next model.

Proprietary data is illiquid: hard to exchange, collect, and sell or license (legal liabilities and other bottlenecks). This scarcity increases the market value of such data.

This paper asks a simple question: What would it take to make private data trade as a commodity?

1.1 The answer rests on three pillars:

Proof of where the data came from (provenance)
Proof that it follows the rules (compliance)
Proof that it is good enough to use (quality)

2. How commodity markets became liquid in the first place

2.1 Oil and wheat are liquid for a few simple reasons:

Grading: A barrel of Brent crude is defined by where it is from and how pure it is. Buyers know exactly what they are getting without checking every barrel by hand.
Two separate markets: The spot market sells the real thing (physical goods); the futures market trades a contract on a future price of that real thing. Most futures contracts are never delivered (~2% settle), and the futures market for oil is ~30x the size of the spot market.
Scale: Because the contracts are standard, the same barrel of oil can be referenced by many traders at once. The ratio of paper trading to physical trading has grown sharply since the 1990s³, and approximately 1-5% of futures contracts ever result in physical delivery.⁴

Data could follow a similar pattern. But data has a problem oil doesn't have: it can be copied infinitely, arguably for free.

3. Why proprietary data is harder

Generic data, like weather readings and stock prices, is already close to a commodity.

Proprietary data, as stated in the introduction, belongs to some entity. Selling it without that entity's permission is often illegal, and always a trust problem.

3.1 Copying breaks scarcity

Once a buyer has the data, they can copy it endlessly. A seller loses control the moment it changes hands.

3.2 Re-identification is a hidden risk

Even data that looks anonymous can sometimes be traced back to a real entity once combined with other data, as researchers did by re-identifying people in the anonymized Netflix Prize dataset (Narayanan & Shmatikov, 2008)⁵.

This makes proprietary data riskier to trade than a barrel of oil.

So any plan to liquidate proprietary data must solve three problems at once: where it came from, whether it follows the rules, and how good it is.

4. The Three Pillars

4.1 Provenance

Provenance means proof of where the data came from and who agreed to share it. It should answer:

Who collected this?
When?
Did the entity consent?
Has it been changed since?

Without this proof, a buyer cannot trust the data, and a seller cannot prove they have the right to sell it.

A web standard called PROV already defines how to record this kind of history for digital data.⁶

4.2 Compliance

Compliance means the data follows privacy law and the rules of consent.

It should answer:

Is this data allowed to be sold in this form?
Has it been anonymized correctly?
Does it meet the rules of the places where it will be used?

Compliance turns a legal risk into a known, checked fact. Under GDPR, for example, data stays regulated as personal data until it is anonymized so that no individual can be identified.⁷

4.3 Quality

Quality means the data is accurate, complete, and current enough to be useful. It should answer:

How much is missing?
How old is it?
How was it measured?

Quality is what makes data worth paying for, not just safe to buy.

Together, these three pillars do for private data what grading does for oil. They turn something messy and personal into something a buyer can trust without checking it by hand.

5. Mechanisms that could make this work

Mechanism	Pillar it serves	What it does
Data passport	Provenance	A standard record attached to the data covering source, consent, and change history
Consent ledger	Provenance, Compliance	A timestamped log proving an entity agreed to share their data, and for what use
Differential privacy	Compliance	Adds controlled noise to numbers so no single person can be identified
Synthetic data	Compliance	Creates artificial data that behaves like the real data, without exposing real people
Clean rooms	Compliance, Quality	Lets a buyer run analysis on the data without ever holding a copy
Independent data audits	Quality	A third party checks accuracy, completeness, and freshness, like a grain inspector
Tokenized access	Delivery	Sells a time-limited right to use data, not the data itself
Cryptographic watermarking	Delivery	Marks each copy so a leak or resale can be traced back to its source

No single mechanism does the whole job. A working market would combine several:

Data passport for provenance
Differential privacy or synthetic data for compliance
Independent audits for quality

Delivery would lean on access, not transfer, since private data should rarely leave a safe, controlled environment.

These are not hypothetical. The US Census Bureau used differential privacy to protect the 2020 Census. ⁸

About two-thirds of US data and advertising professionals already use clean rooms. ⁹

And Gartner has predicted synthetic data would make up most AI training data by the late 2020s.¹⁰

6. What a liquid market could look like

A private data exchange would work like a regulated marketplace, not a free-for-all.

Marketplaces for generic data already exist. AWS Data Exchange alone lists more than 3,500 third-party data sets. ¹¹

A private data exchange would need the same kind of catalog, plus the consent and compliance layers described above.

Listings would show a data passport instead of the raw data: source, consent status, compliance grade, and quality score. Buyers would see this before any data moves.

Access, not ownership, would be the product. Most deals would be time-limited rights to query or train on data inside a clean room, not a file transfer.

An escrow and clearing layer would hold payment until an independent audit confirms the data matches its listed grade. This removes the need for buyers and sellers to trust each other directly.

Over time, reference prices could emerge for common categories, such as verified health records of a certain type, or anonymized transaction data from a certain sector.

Once prices exist and move, derivatives become possible: contracts that let a buyer lock in future data costs, the same way an airline locks in future fuel prices.

7. Risks and limits

Liquidity is not automatically good. A few risks need honest attention.

People are often left out of the value they create. A person's data may get sold many times while they are paid once, or never. Fair compensation models are still unsolved. Economists have proposed treating data as a form of labor people should be paid for, though no large-scale version of this has been built yet.¹²

Re-identification risk never fully disappears. Even well-anonymized data can sometimes be reversed with enough outside information. No grading system removes this risk completely; it only reduces it.

Standardization can strip out value. Forcing data into a fixed grade may remove the very detail that made it useful. Not all private data should be commodified. Some should stay closely controlled, deal by deal.

These limits do not rule out a market. They mean it should start narrow, in categories where consent, risk and rules are already clear.

8. Conclusion

Private data is not liquid today because it fails every test a commodity must pass. There is no standard grade, no easy way to check quality, no safe way to deliver it, and no clear proof of consent.

The path to liquidity runs through three pillars: verifiable provenance, compliance, and quality. Each needs its own tools:

Data passports
Consent ledgers
Differential privacy
Clean rooms
Independent audits

A full market for all private data is not realistic soon. A narrow market, built one well-defined category at a time, is. Health records, financial transaction data, and location data with clear consent are the most likely starting points.

If this path works, private data will not trade like oil overnight. But it can become something closer to a modern, regulated asset:

Tradable
Auditable
Fair to the entities it comes from

References

Author — https://mohamedtah.com ↩
Voidback — https://voidback.com ↩
Oil Futures Market / Spot Market — https://economymiddleeast.com/news/virtual-barrels-dwarf-trade-physical-oil-prices/ ↩
Oil Futures Settlements — https://phillipcapitaldifc.ae/futures-settlement-methods/ ↩
Netflix Dataset Prize — https://arxiv.org/pdf/cs/0610105 ↩
W3C — https://www.w3.org/TR/prov-overview/ ↩
Data Protection Commission Ireland — https://www.dataprotection.ie/en/faqs/general/what-personal-data ↩
2020 Census — https://arxiv.org/pdf/2204.08986 ↩
Data Cleaning Rooms — https://bidscube.com/blog/what-is-a-data-clean-room/ ↩
Synthetic Data — https://www.techmonitor.ai/?p=399955 ↩
AWS Exchange — https://aws.amazon.com/data-exchange/ ↩
Is Data Labor? — https://aeaweb.org/doi/10.1257/pandp.20181003 ↩