The intangible estate: why judgement, values and voice are the assets your will cannot carry

• In 2024, researchers at Stanford and Google DeepMind built AI agents of 1,052 real people from two-hour interviews; the agents reproduced the people's own survey answers with 85% of the accuracy those people showed when re-tested themselves two weeks later (Park et al., 2024).

• The EU Artificial Intelligence Act requires AI-generated or manipulated likenesses of people to be disclosed as such, with the obligation applying from August 2026 (Regulation (EU) 2024/1689, Article 50).

• Tennessee's ELVIS Act, in force since July 2024, is the first US state law to protect a person's voice as a property right against unauthorised AI imitation, during life and after it.

Ask anyone who has administered an estate what the family wanted from them most, and it is rarely the accounts. It is a question the documents cannot answer: what would she have wanted? The house, the business, the grandson who wants to leave university, the sister who never called; every one of them arrives at the executor's door as a question about a person, and the executor has only paper.

The paper is good. Over centuries the law has built an extraordinary set of instruments for passing on what a person owns and for controlling how it is used after they are gone. What it has never built is an instrument for passing on the person: how they weighed things, what they would never do, what they would say to a daughter about an offer on the table. That part of an estate has always been left to memory, and memory is the one asset that depreciates from the day of the funeral.

We call it the intangible estate, or the human estate: the third part of everything a person leaves, beside the financial and the legal. This article is about whether it can be captured, who would own it if it were, and what a family should insist on before anyone tries.

It is worth being precise, because the industry's instruments are better than most families realise, and their limits are exact.

Read the right-hand column as a list. It is the intangible estate, item by item: the why, the how-to-choose, the unforeseen case, the decision itself, the purpose of the money, the voice in the room. The letter of wishes is the closest the law has come to holding any of it, and it is a single private page most trusts never receive.

The intuition most people have is that a person is not the kind of thing that can be captured, and the intuition is partly right. What can be captured is narrower and more useful than the science-fiction version: not consciousness, not the person, but the person's manner of answering, deciding and explaining, well enough that the answers match.

The clearest evidence is a 2024 study by researchers at Stanford University and Google DeepMind. They interviewed 1,052 people, chosen to reflect the US population, for about two hours each, in the person's own words and at the person's own pace. From each interview they built a generative agent, a language model given the transcript and asked to answer as that person. Then they tested the agents against the people. On the General Social Survey, a standard battery of attitude questions, the agents reproduced their person's answers with 85% of the accuracy the people themselves achieved when asked the same questions again two weeks later. (Park and others, Generative Agent Simulations of 1,000 People, 2024). On personality inventories and on economic-game behaviour the match was weaker but still well above chance, and the agents were less biased across racial and political groups than the usual alternative, which is to predict people from their demographics.

Three things in that result matter for families. The substrate was the person's own words, not a questionnaire about them: two hours of a person talking about their life outperformed anything built from a profile. The interview was open, not steered; the interviewer followed the person. And the ceiling was set by the person's own consistency: nobody answers the same way twice, and the agents came close to as consistent as the people were with themselves. The limit is real. An agent built this way does not know what the person never said, does not grow the way a living person does, and cannot be asked about a future the person never saw. What it can do is answer, in the person's manner, the kind of question an executor gets asked.

That is the whole claim, and it is enough. The intangible estate is not a ghost. It is a body of the person's own words, held in a form that can be asked questions, that answers the way they answered.

The law on this is young, uneven, and moving quickly, and anyone building or buying such a thing should know where it stands. The following is the position as at September 2026; each line is confirmed by the reviewer against the statute before publication.

Three questions run through every row. Consent: did the person agree, while living, to a likeness being made, and to what use? Ownership: whose property is the likeness and the recordings it is built from? Control: who can change it, stop it, or switch it off once the person is gone? Any family considering this should be able to answer all three in writing, and any company offering it should answer them before being asked.

Here is how we answer them, as facts about the product rather than promises about the future.

A Personal AI Twin on Timeless AI is built from what a person says in their own words: the stories they tell, the memories they record, the questions they answer, and their voice, recorded by them. Nothing is built from anyone else's account of them. While the person lives, they hold it: they add to it, they hear it, they decide who among their people may reach it, and they can end it.

At the moment the person chooses, the Executor Lock™ closes. The Persona is frozen as a complete snapshot: every memory kept, none pruned or reweighted, the personality fixed from that day forward. The people the person named, their executor and trusted contacts, hold the key. Events can still be added around the edges, so that the Persona can know what has happened since, but who the person is never changes again. That is the whole design principle: a likeness that cannot drift into someone the family did not know.

The recordings and the Persona belong to the person. The Executor Lock™ Agreement defines the Persona as "the AI representation of the Persona Owner, including all associated memories, media, emotional metadata, and governance settings", and the Persona Owner as "the individual who created the Persona". After the lock, governance passes to the Executor the person named, and the Trusted Contacts they chose reach the Persona within the permissions the person set; the Executor Lock™ terms, accepted by both the person and the Executor before the lock exists, say exactly that. It is disclosed as what it is, on every surface, every time: a Persona, built from the person's own words, never the person.

Put beside the table above, it is the seventh row: the instrument for the right-hand column. It does not replace the will, the trust or the power of attorney. It sits beside them, the fourth document, and carries what the other three were never built to carry.

The research on why wealth transfers fail, set out in The Great Wealth Transfer, explained for families, finds that the failures are overwhelmingly human: trust breaks down, communication stops, heirs are unprepared. Read through the three-estate model, every one of those is the absence of the person at the moment the person was needed. The heirs did not lose the money first. They lost the judgement that had held it together, and the money followed.

For an adviser, the practical consequence is simple. The moment a principal dies is the moment the adviser's relationship with the family is most at risk, and the reason is that the family is now making decisions the principal used to make, without the principal's reasons. An adviser who can put the principal's own account of those reasons in the room, in the principal's own words, is a different kind of adviser. For an insurer, the logic is the same one step earlier: the policy pays out the money; the Persona answers the questions the money raises.

These are commitments, not disclaimers. They bind the company.

1. Consent while living, or nothing. A Persona is built only by the person it is a likeness of, from their own recordings, while they are alive to choose. We do not build likenesses of people from what others remember of them.

2. No invented memories. A Persona speaks from what the person said. Where the person said nothing, it says so.

3. The Executor can close it, and nobody else can change it. After the lock, the Persona is held frozen for the retention period in the Executor Lock™ terms. We do not modify, retrain or supplement it, we do not give third parties access to it, and we do not delete it, except on the Executor's written instruction. Those words are in the agreement the person and their Executor both accept before the lock exists.

4. Always disclosed. A Persona is presented as a Persona on every surface. It never claims to be the person.

5. Nothing sold to grief. We do not advertise to people in the days after a death. What a family pays is set out before anyone dies, in the plans and in the Executor Lock™ terms.