Your firm adopted AI months ago. Nobody decided to

On an evening in the winter of 1881, in the Long Gallery at Hatfield House, sparks fell from the ceiling onto the guests below.

The gallery runs a hundred and seventy feet along the south front of the house. The ceiling above the guests was covered in gold leaf, laid there by an earlier Marquess who had admired a gilded ceiling in Venice and wanted the same effect at home. Along the beams above that gold ran a length of bare copper wire, carrying current from a dynamo driven by water at the mill, feeding a row of new incandescent lamps.

The wire had begun to burn.

Nobody sent for help. The family who lived in the house had been through this before. Cushions came up off the chairs and were thrown at the ceiling, hard enough to knock the flame out of the smouldering patch, and when the burning stopped the cushions were retrieved and the conversation carried on. The account survives because Salisbury's daughter later wrote the family's life down. She recorded the throwing of cushions as an ordinary domestic event, of no more consequence than a door left open in a draught.

Robert Gascoyne-Cecil, third Marquess of Salisbury, was then the leader of the Conservatives in the Lords, and would be Prime Minister within four years. He was also, by inclination, an experimenter. He kept a laboratory in the house. He had taken an interest in Joseph Swan's lamps almost as soon as Swan had a lamp worth taking an interest in, and he wired Hatfield himself, with his sons and members of his household, running cable through a building finished in 1611.

The building mattered. Hatfield is oak panelling, oak beams, deep voids behind the walls, and floor cavities nobody had opened in two centuries. Everyone in the house understood what fire did to such a place, because the house had already burned. In November 1835 the west wing went up in the night. The Dowager Marchioness, then eighty-five, was believed to have caught the feathers of her headdress alight at her writing desk. The wing was destroyed. Of the Marchioness, almost nothing was recovered. Servants living at Hatfield in 1881 had been born into a household that told that story.

So the current running through the beams was not an abstraction to anyone there. And yet the way the current arrived was casual almost past belief.

There was no plan of the installation. There was no drawing showing which beams carried cable and which did not. Joints were twisted by hand and wrapped where wrapping seemed sensible. Insulation, where insulation existed, was cotton and gutta-percha, and both perished unseen in warm cavities where nobody looked. The work was done by people who cared about the house and were proud of the light, which is exactly why they kept doing more of it.

The light spread the way water spreads. A lamp in the laboratory, then lamps in the gallery, then a fitting in a corridor because carrying a candle along that corridor had always been a nuisance. Each addition was small. Each addition was reasonable. Each was made by somebody competent and well-meaning who had no reason to write anything down, because the previous addition had not been written down either, and nothing had gone wrong that a cushion could not settle.

Nobody in the house was concealing anything. There was nothing anyone thought worth concealing. The men who ran the cable would have described their work to any visitor who asked, in detail, with pride. The trouble was that no visitor asked, because the question that produces a useful answer is a strange question to think of asking. Nobody walks into a house and enquires how much of the building is carrying current. The absence of the question, rather than any wish to hide, is what kept the installation off every record the household kept.

By the early 1880s no person alive could have stood in the entrance hall at Hatfield and said, with accuracy, how much of the house was live. Not the steward. Not the staff who had run the cable. Not Salisbury.

The reckoning, when the reckoning came, arrived from outside and arrived as paperwork. Fire insurers had begun to notice that the houses and theatres taking up electric light were burning at a rate that troubled them. In 1882 the engineers responded with the first published rules for wiring, drafted under pressure from the insurance offices, setting out how cable should be run, jointed, insulated and protected.

Every rule in that document rested on a single assumption. The assumption was that somebody could say what was already in the walls.

At Hatfield, as at every other house that had electrified itself by enthusiasm, that question had no answer. The rules could be applied to the next hundred feet of cable. The four years of cable already installed sat behind the panelling, in a house nobody had ever surveyed, doing whatever heat and time had persuaded the cotton to do.

Somebody walking out onto the gravel at dusk in that summer of 1882 would have turned and seen the south front give back brick, stone mullions and the roofline four Marquesses had inherited, a face unaltered since 1611, no scaffold against the stone, no scar in the brickwork, no day marked in any diary and no minute taken in any book, while behind a window on that same front, in a room that appeared on no list, above beams no drawing recorded, fed by cable jointed by hand and forgotten by everybody who had ever touched the cable, a filament burned.

The light was already on inside.

A firm can sit in the same condition, and many do.

Ask the founder of a UK advice firm whether the firm has adopted AI and the answer is often no. The answer is given in good faith. No budget has been approved, no supplier has been signed, no policy has been circulated, no announcement has been made. Measured against everything a founder can see from the front of the house, the answer is correct.

The answer describes the front of the house.

Inside, a paraplanner is pasting the bones of a suitability report into a chatbot to sharpen the wording, because the wording takes an hour otherwise. An adviser is running a recorded client meeting through a summariser to save an evening. Somebody in operations has found a tool that turns a spreadsheet into a client-friendly explanation and has started using the tool on Fridays without mentioning the fact. Each of those people is diligent. Each is trying to do the work better. None of them is hiding anything, because none of them believes there is anything to hide.

Nobody decided. Which is exactly how the light spread through Hatfield.

Firm size changes the shape of this and not the substance. In a practice of four, the person using the tools most often is the founder, drafting on a Sunday evening and thinking of the work as writing rather than as technology. In a firm of forty, the founder may be the only person in the building who has never opened a chatbot, which makes the front elevation more convincing and the wiring harder to see.

The numbers around this are worth holding lightly and taking seriously. The Bank of England and the FCA surveyed financial services firms and reported in November 2024 that three quarters of respondents were already using AI, with a further tenth planning to start within three years. Two years earlier the same survey had found well under two thirds. In June 2025 the FCA's chief executive put adoption above eighty per cent and moved the regulator's stated concern away from encouraging firms to start and towards helping firms deploy at scale.

Those figures cover firms, and firms are the easy part to count. The harder question is what employees are doing without being asked. Microsoft's UK research in October 2025 found seventy-one per cent of employees had used AI tools their employer had not approved, and half were doing so every week. Separate surveys through 2025 and 2026 from security vendors and workplace researchers land in a similar band, roughly half to two thirds, depending on who is asked and how the question is worded.

Treat that second set of numbers as a direction rather than a measurement. They are self-reported, they come mostly from vendors with an interest in the answer, and none of them isolates UK financial advice firms. What they establish is not a precise percentage in your firm. What they establish is that a firm with no AI policy is not therefore a firm with no AI in the walls.

The regulatory position removes the last comfortable reading. The FCA has said, repeatedly and in writing, that no separate AI rulebook is coming and that existing requirements already cover what firms do with these tools. Examples of good and poor practice are expected later this year. The practical meaning is the same as the meaning of those 1882 wiring rules. The standard applies to what is in the walls today, not to the installation you have yet to begin.

Which leaves a choice of venue for the discovery, and only two exist.

The first venue is a review, an audit, a complaint, or a data incident. In that room the finding arrives with a confession attached, and every question afterwards is about what else has been missed.

The second venue is a Monday morning of your own choosing. The finding is identical. The framing is yours. Nothing has gone wrong, nobody is being investigated, and the firm is doing a piece of fact-finding that any sensible business would do about any tool in daily use.

If you run an advice firm, the useful work looks like this.

Survey before you legislate. Send an anonymous, no-blame questionnaire to everyone: which tools, which tasks, which client information, how often. Say in the covering note that nobody is in trouble, and mean that. Give people a week.

Credit the people who answer. The staff using these tools are usually the ones who care most about the quality of the output. Treat the disclosure as initiative, because blame produces a quieter survey rather than a smaller problem.

Write down what comes back before deciding anything. A single list of tools and tasks is the first accurate description of your firm's position that has existed. That list is the asset. The policy is downstream.

Name the owner. A record without an owner rots inside a quarter. Somebody senior holds the list, and the list returns to the table on a date already in the diary.

Set your own line, and expect the line to move. What passes for acceptable use in a firm with in-house paraplanning fails in a firm that outsources, and the same tool can be fine for a first draft and wrong for anything a client sees. That call belongs to your compliance officer and to you. A system should make the position visible and checkable. The judgement stays with the professional making it.

None of that is a project. The survey costs a week and the answer is yours to keep.

The house at Hatfield stood. Salisbury lived another twenty years, served three times as Prime Minister, and kept his lamps. The family were careful people who loved the house. The gap that mattered ran between the front elevation and the wiring, and closing that gap required somebody to ask a question nobody had thought to ask, about a system nobody remembered deciding to build.

Your firm has a front elevation too. The website, the policy folder, the last supervision file, the answer you would give a regulator this afternoon about what technology the firm uses.

Somebody in your building could describe the wiring in detail, and with some pride. Nobody has asked them yet.

The questionnaire that asks them fits on a single page.