Parliaments, councils, government agencies and public institutions. Hansard, committee proceedings, and the systems that produce the official record — where accuracy and accountability are not features but obligations.
Artificial intelligence is the largest change to reach the production of the civic record in a generation. This paper is a set of observations from inside the work: where AI genuinely helps, where it must not be allowed to, how the people who keep the record carry it through the change, and the question of sovereignty that sits beneath it all — who, in the end, controls the record?
Paper 1 made the case for AI on the mechanical side of producing the civic record. This paper says what holding that line actually requires, given a structural condition the first paper did not state plainly enough: the workforce trained to produce the record from raw audio is contracting everywhere. AI adoption is no longer optional. The question is what disciplined adoption requires — and what sovereignty means when the institution depends on tooling not just to keep records, but to make new ones.
If a Hansard organisation were to design its production environment for the AI era from first principles — rather than retrofit AI onto an inherited workflow — what would the architecture look like? This paper sets out a design for a system that treats the reporter's live log as the foundation, separates the live captioning obligation from the considered record, places AI agents only where human judgement is consultatively supported, and produces an open-source XML archive as the long-term preservation artefact. Working draft of the architecture is in preparation.