Monthly Register — February 2026

The question came up in a progress meeting, put by the client's project manager, and it was the right one: if our designer's parent now owns the tool that has read our model, whose model is it? Nobody in the room had an answer that fitted on a slide.

What moved

February's movement is a class of purchase rather than a named close. Professional-services houses that already hold the appointments — design, project management, advisory — buying the AI layer that was sold as sitting on engineering. The trade press will print a consideration when it has one. The number is seldom the interesting part. What matters is what the purchase does to the information the appointment was already producing.

Follow the ordinary chain. A client appointed a designer. The designer produced a project information model. That model lived, in practice, across a stack the designer did not own: a hosted common data environment, an authoring licence, and somebody else's retrieval layer sitting on top of both. After a close of this type, the last of those belongs to the firm that already holds the appointment. Adviser and platform stop being two companies and become one commercial object with two invoices.

That is not automatically a failure. It can be a cleaner stack — fewer contracts, one support line, one party to chase when the export breaks. It is a change of custody all the same, and custody is what gets argued about when a job goes wrong.

The questions worth asking on the Monday after are ordinary and surprisingly hard to answer. Which containers were published under the appointment, and to whom. Whether the acquired tool held work in progress, shared material, published material, or a fourth pile that was never really a state at all. Whether training, fine-tuning or retrieval indexes were built on that appointment's files. Whether the appointment, the AI terms, or both, say anything about who owns derivatives. Whether a complete extract can still be taken — by the client, and by every other appointed party, not just the one with the strongest commercial position. Whether other clients' jobs sat in the same index as yours.

A model that retrieves "the latest" from a consultancy-owned stack is a view, not the published set. If that view cannot name a revision and a suitability, what the job has bought is a cover sheet with a new owner.

Leakage across appointments is the unfashionable risk, and the one nobody wants to raise in the first meeting. Engineering-AI houses index, and sometimes train, across a book of work. A global consultancy's book of work is other people's projects. An acquisition does not invent that problem; it concentrates it. The client who asked for a copilot on their federated model may now be a row in a house that also advises the competing bidder, the operator and, in time, the claims consultant. ISO 19650-5 already treats sensitive information as a classification and sharing problem. A press release classifies nothing.

There is a middle tier that tends to get forgotten in all of this. The structural subconsultant, the fire engineer, the specialist façade designer: parties appointed by somebody else, who never held a seat in the retrieval tool, and whose drawings are nonetheless sitting in the index. Nobody writes to them on the Monday after. They find out weeks later, when a query comes back quoting something they issued at stage three and have since superseded twice, and somebody then has to establish whether the index was ever told about the supersession.

Meanwhile the unofficial set carries on regardless. The cabin still has a USB stick in a drawer. The plotter still runs on a Thursday afternoon. What changes is the story told about the official stack, which now looks more unified because one firm sits on both the advice and the retrieval. Unity of letterhead is not unity of state. The clerical labour does not disappear either; it moves to whoever is asked to prove a published container after the licence, the subsidiary or the book of work has changed hands.

This is a different problem from the vendor-cloud argument that went round last autumn. That one was about files living in somebody else's platform. This one is about the professional-services firm producing the engineering coming to own the memory that sat on it. Both end at the same document-control question. Only one of them can be answered by an export clause in a software schedule; the other has to be answered in the appointment, in the information protocol, and in a variation nobody has drafted yet.

What changed in the standards

No new part of ISO 19650 was published in February. An acquisition is not a publication, and a slide claiming the series has moved because a consultancy bought a model is not a part of anything. Somebody at a project meeting will ask, in good faith, whether any of this changes which parts apply to the appointment. It does not, and saying so early saves a fortnight of circular email.

The texts appointments still point at are the published parts already in circulation: Part 1 and Part 2 from 2018, Part 3 from 2020, Part 4 from 2022, Part 5 from 2020, and Part 6 from January 2025. They were not new this month. They were in use this month, which is a different fact and the more useful one.

Those texts assume an appointing party, appointed parties, an information protocol, a common data environment and a published container. They do not assume that an appointed party's corporate group will come to own the retrieval layer sitting on the project information model. They do not forbid it either. What they require is that information requirements, exchange and security are named. If the BIM execution plan still says the common data environment is last year's host, while the assistant now lives inside a house the lead consultant has bought, the plan is describing last year's custody.

The revision of Parts 1, 2 and 3 remained a drafting conversation through February. Committee drafts and comment labour belong to the ISO project record, and they do not acquire a fresh dateline because a purchase closed. Live appointments saying ISO 19650 aligned were still aligning, if they aligned to anything, to the published language. Treating a closed comment period as though the new parts had arrived is how a job buys a slogan and keeps the same register it had before.

The Building Safety Act's information duties in England were already in force. They did not arrive with a deal. They still require an identifiable, available, change-controlled golden thread — not a chat log inside a new subsidiary, and not an index the principal accountable person cannot produce when somebody asks for it.

Gulf permit-stage BIM requirements remain a class of public-record rule where the text is public. If a specific annex, circular or draft ballot moved in February, it belongs in a later issue once it can be cited rather than in a reconstructed paragraph here.

The standards point that actually belongs in this month is about ownership as written rather than ownership as hoped. Who is the appointing party. Who is appointed. What is published. What is security-classified. An acquisition moves the corporate owner of a tool; on its own it moves nothing in the protocol. Treating those as the same sentence is how a job discovers, at handover or in a claim, that the memory of the work sits in a licence the client never held.

One thing to watch

Watch the word "ours" in the months after a close. If it means the published set the appointing party can extract tomorrow, fine. If it means the assistant the consultant now owns, somebody has bought a rumour and will find out which at the worst possible moment.

Watch whether the move from project information model to asset information model is still nobody's job now that the project model's working memory has changed group. Handover does not get cleaner because the same letterhead sits on the advice and on the index. It turns into a question about whether the operator inherits a published container or a licence they were never party to, and operators are usually the last people in the room to be told.

Watch what the assistant does with superseded material, which is the dullest test available and the most informative. Ask it something you already know the answer to, on a package where a revision was withdrawn, and see which sheet it quotes back. If it answers confidently from the withdrawn one, the index is not doing the work of a register, and no amount of corporate tidying upstream will fix that.

Three things are worth putting in writing, and the appointing party is the only one who can sensibly ask for them. A one-page custody position: which systems hold project information today, who owns each of them, and what happens to the index if the corporate owner changes again. A tested extract rather than a promised one, taken by somebody other than the incumbent, opened in another tool and checked against the register. And an honest answer on whether the information protocol needs a variation, because it usually does, and a variation costs money and takes the better part of a quarter.

Two weeks after a close the slides have gone quiet. If document controllers are still arguing about who can name the issued set, that is a filing problem wearing a mergers headline — and it will still be sitting on somebody's desk at handover.