What a data trust actually is
The structure behind the phrase, built one part at a time, and how it differs from a cooperative, a commons, a marketplace or a contract.
Sample article: the structure is final, the copy is a first draft to replace with yours. Every source it cites is linked.
The phrase does two jobs at once, which is why it causes arguments. The 2017 review of AI in the UK by Wendy Hall and Jérôme Pesenti put "data trusts" into policy, meaning repeatable frameworks and agreements for sharing data between organisations — not a legal entity. The Open Data Institute, which ran pilots from 2018 with the Office for AI and Innovate UK, means something narrower and harder: "legal structures that provide independent stewardship of data".
Those are different claims. One is a template for contracts. The other borrows a 500-year-old idea from the law of trusts, where someone holds a thing for someone else and owes them a duty for as long as they hold it. This article is about the second one, because the second one is the one that changes who has power.
The structure, a part at a time
Each part exists to close a specific hole. Step through them and the reason for the shape becomes clear: it is built to survive the trustee changing their mind.
Rights, not files
A trust is useless if it only holds copies. What it has to hold is the authority to decide — otherwise whoever kept the original can still do as they like.
Contributors transfer the rights they hold over data — to grant access, to set terms, to say no — rather than handing over the data itself. The files may never move.
A pile of data with no settled authority over it. Every request reopens the question of who may say yes.
The key move is at step one. A trust does not need the files — it needs the rights over them. Data can sit exactly where it is, in the systems that already hold it, while the authority to say yes or no moves somewhere independent. That is what makes the arrangement possible at all for data nobody is willing to copy.
One answer among several
A trust is not the default and not the goal. It is one answer to the question who decides what may be done with this data, and it is a heavy answer: it needs a trustee, a purpose, and someone to hold the trustee to it. The Ada Lovelace Institute and the AI Council put trusts alongside cooperatives and plain corporate and contractual models for exactly this reason — the mechanism should follow the problem.
An independent trustee holds the rights and exercises them for named beneficiaries.
- Holds the rights
- The trustee, on the beneficiaries’ behalf
- Decides
- The trustee alone
- Bound by
- A fiduciary duty of loyalty and care, enforceable by the beneficiaries
- Fits
- Many contributors with weak individual bargaining power, and decisions too frequent to put to a vote each time
You are swapping one concentration of power for another. Everything rests on the trustee being, and staying, independent.
Where it stands
The pilots the ODI ran — on food waste, the illegal wildlife trade, and city data with the Greater London Authority and the Royal Borough of Greenwich — were deliberately exploratory, and the honest summary is that data trusts remain more argued about than built. The Data Trusts Initiative is where the legal work continues.
The idea is worth understanding anyway, because it names something most data sharing lacks. Almost every arrangement in practice binds the people who hold data to each other. A trust is an attempt to bind them to the people the data came from.