The problem: the people who make the data rarely own it

Scientific progress runs on data — genomes, clinical measurements, sensor streams, browsing and behavioural traces. Yet the people and institutions who generate that data almost never control it or share in the value extracted from it. A research participant hands over a sample and gets a one-time payment, if anything; a patient's records are locked inside a hospital system; a consumer's DNA sits on a company's servers under a terms-of-service agreement they cannot renegotiate. The value accrues to whoever holds the database.

What that concentration means in practice was made vivid in 2025, when 23andMe Holding Co. and certain subsidiaries filed voluntary Chapter 11 petitions on 23 March 2025 and the customer genomic database became an asset of the estate. The company's own annual report for the year ended 31 March 2025 sizes it at “approximately 14.4 million and 15.1 million PGS customers” for 2025 and 2024 — a fall it attributes “primarily as a result of customers electing to delete their data” — and reports that 80% of customers had consented to research. It was auctioned in May 2025 among seven qualified bidders, and after a reopened round closed on 14 July 2025 at $305.0 million to TTAM Research Institute — a nonprofit affiliated with the company's co-founder and former CEO — “free and clear of liens, claims, encumbrances, and other interests.” It is the sharpest illustration of the core problem decentralised science takes aim at: when data ownership is centralised, the terms can change without the consent of the people the data describes, and the only lever an individual holds is deletion. See GenomesDAO for a vault design built specifically against this failure mode.

Data as an ownable, governable asset

The web3 answer is to represent a dataset as something you can hold, permission, and trade without giving up custody of the underlying records. Ocean Protocol's datatokens are the clearest example: each published dataset is paired with an ERC‑20 datatoken that acts purely as an access-control key, while a data NFT represents the base ownership right. Holding and spending the datatoken grants access to the data; the data itself never has to go on-chain and can stay on the owner's own storage. Ownership, licensing, and access all become programmable, which is precisely what lets a group of people pool and govern data collectively.

This turns two things that were previously trapped inside a single company into shared, tradeable objects: the right to the data and the revenue from licensing it. It is the data analogue of the IP-NFT pattern DeSci already uses for patents and research IP — ownership of an intangible asset made liquid and collectively holdable.

Data DAOs and data unions in practice

A Data DAO is a collective that owns or manages data assets on behalf of its members, pooling many individually-worthless contributions into a dataset with real market value and governing the terms of access together. Several distinct models exist:

• Ocean Protocol — the base layer for tokenised data. Its compute-to-data mechanism lets a buyer run an algorithm against a private dataset and receive only the result, so a DAO can monetise sensitive data for AI training without the raw records ever leaving the owner's server.
• GenomesDAO — genomic-data ownership built on encrypted-vault infrastructure (AMD SEV hardware enclaves). Individuals keep their sequenced DNA in a private vault and are paid each time a researcher queries it, rather than selling it outright to a sequencing company.
• Data unions (Streamr) — opt-in pools of real-time personal data where a smart contract splits buyer revenue among all contributors, originally paid out by Streamr's Monoplasma distributor. The pattern outlived the vendor that named it. That defining post calls a data union “a framework, currently being built on the Streamr Marketplace” — a surface that now redirects to a generic project hub — and Streamr's live documentation is entirely pub/sub, nodes and staking: a census of the 120 URLs in docs.streamr.network/sitemap.xml on 9 September 2026 returns no page matching “union” at all. Streamr paused fundraising for the Data Union DAO on the grounds that “the business case for a dedicated Data Union infrastructure provider has not been strong enough to attract venture capital to date”, and carried the payout engineering over into Rail, “replacing Data Union DAO and UNION for Rail and the RAIL token”. The consumer end kept running without it: Swash is live and still pays people for their data, but presents itself as “an online earning portal where you can earn points for being active and completing tasks online” rather than as a data union, and publishes no member count. Read the model, then, as a design that works and a dependency that did not: the revenue-splitting contract is the easy half, and the entity willing to maintain it is the hard one.
• AxonDAO — the consumer-biometric variant: members contribute health metrics, behavioural signals, and voice recordings and are paid in $AXGT for consented access, with the raw records kept off-chain in compliance-ready storage and only the access rights coordinated on-chain.

The common thread is the same one that defines a DeSci DAO: a shared asset (here, the data and the revenue it earns) owned by many, with the rules of access decided by governance rather than by a single custodian.

The hard problems

Tokenising data does not make the underlying difficulties disappear, and honest coverage names them:

• Privacy and re-identification. Genomic and health data can never be truly anonymised — a genome identifies not only you but your relatives. Compute-to-data and hardware enclaves reduce exposure but do not eliminate the risk that a determined buyer re-identifies contributors from query results.
• Valuation. A single person's data is nearly worthless; value only emerges at aggregate scale, which makes it genuinely hard to price any one contribution fairly and easy for a marketplace to under-reward the people supplying the raw material.
• Regulation. Selling or transferring personal data collides with GDPR, HIPAA, and a growing patchwork of state genetic-privacy laws — some of which, as the 23andMe objections showed, treat consent as non-transferable regardless of what a token says.
• Governance capture. If control of a data collective is a freely-bought token, an outside buyer can accumulate it and outvote the very contributors whose data gives the pool its worth — the classic DAO failure, applied to data.

That last hazard is not hypothetical, and the sharpest case is the base layer named at the top of this section. Ocean's token was folded into the Artificial Superintelligence Alliance alongside Fetch.ai and SingularityNET — an arrangement Ocean itself describes as having been “founded on voluntary association and collaboration to promote decentralized AI through a token merger.” On 9 October 2025 the Ocean Protocol Foundation withdrew its designated directors and resigned as a member, and the same notice put numbers on what leaving costs the people who followed the token in: since July 2024 81% of the $OCEAN supply had been converted into $FET, 37,334 holders representing 270 million $OCEAN had not converted, and the conversion bridge was “fully managed and controlled by Fetch.ai” – the counterparty, not Ocean. Ocean's own account of how it got there is long, partisan and contested. The data stack is unaffected – datatokens, data NFTs and compute-to-data were all still the live product when this page was checked on 9 September 2026 – but the token that pays for them spent eighteen months governed by an alliance its own foundation then walked out of. DAO mergers and token consolidations covers the mechanics and what a dissenting holder is actually left holding.

Where DeSci — and Caper — fit

Data ownership is the third leg of the DeSci agenda alongside funding and IP: fund the work, own the resulting IP, and let the people who generate the data keep and govern it. The data-custody layer — datatokens, compute-to-data, encrypted vaults — solves how a dataset is held and accessed. It leaves open a second question that is pure governance: once a collective owns a valuable pool of data and the revenue it earns, who decides the terms, and can a contributor who disagrees leave with their share?

How Caper approaches this. Caper is a DAO protocol, not a data-storage layer — it would sit underneath a data collective as the funding-and-governance rail, next to (not instead of) an Ocean- or GenomesDAO-style custody stack. Its relevance is to the last problem above. A caper's decisive governance weight is (t · v) / (V · T) — your token share multiplied by your share of an earned, non-transferable soulbound vote token. That record is minted on two surfaces at one flat rate – one token per ranked ballot (a 100 XRD vote fee at 0.01 v per XRD) and 0.01 v per XRD of gross value on every buy and every sell – and it is soulbound in the strict sense: it cannot be bought from another holder, sent, or delegated. So control cannot be lifted off a seller in a single block. An outside buyer can still accrue the record, but only by running their own capital through the curve at the price their own buying moves, or by turning up to vote. That prices capture rather than forbidding it. That earned factor maps naturally onto the contribute-to-earn logic data unions already use. And every member holds a clamped exit right: a contributor who dislikes where the collective is taking its data can redeem their share of the treasury and leave, rather than being trapped in a pool whose direction they lost a vote on.

Two honest caveats. Caper governs the shared treasury and decisions, not the data itself — the privacy and custody guarantees still come from the data layer, not the DAO. And its weight is not purely earned: t is a genuine capital term, so a large holder still carries real weight — the earned vote token bounds capture, it does not abolish the influence of size. See also Rejuve.AI, an AI-longevity data union that rewards members for contributing health data.