AI dependence has a number attached to it now. LexisNexis surveyed 543 UK legal professionals for a report published on 2 September 2026, and one in ten of them said they are dependent on AI to do their jobs. Three quarters use it at least once a week and a third use it every day. That describes a profession which has stopped trialling the technology and started leaning on it.
The findings worth your attention sit further down the page. Three quarters of those surveyed think junior lawyers who use AI heavily will struggle to develop legal judgement. Eighty three per cent remain worried about inaccurate or fabricated information, a concern that has not softened as use has spread. And 77 per cent believe AI is leading clients and the public to misread the law, which is a problem arriving at your reception desk rather than one confined to your own drafting.
What dependence looks like in a small firm
In a firm of eight, nobody announces that they have become dependent. It shows in the fee earner who cannot begin a piece of work without opening the tool first. It shows in a file where the prose is smooth and the reasoning is thin, because the model supplied the first and nobody supplied the second. It shows most plainly when you ask where a paragraph came from and get a shrug.
Test for it directly. Take the last three advice letters that went out and ask whoever wrote them to explain, with the tool closed, why the advice lands where it does. Somebody who can do that is using AI. Somebody who cannot is being used by it. The difference matters when a client challenges that advice eighteen months later and the file has to carry the reasoning without its author in the room.
The junior lawyer question
Trainees and paralegals build judgement by producing work that is not good enough and being told why. Remove that and you get people who turn out competent-looking output for years without ever forming the instinct that says something here is wrong. Three quarters of the profession sees the risk. Few firms have changed anything, because the fix costs supervision time now and repays it only later.
Keeping juniors away from the tools is the wrong answer, since it trains them for a job that will not exist. Set the order of work instead. On the tasks that teach most, being a first piece of research, a first advice, a first witness statement, the junior drafts unaided and runs the tool over the draft afterwards to compare. The comparison is the lesson. Where the model caught something they missed, they learn. Where it invented something, they learn faster, and they learn it in your office rather than in front of a district judge.
Then write that sequence down, because a habit living only in a supervising partner's head goes on holiday when the partner does. One page naming which tasks are drafted unaided first, who checks, and what gets recorded, sits comfortably alongside the training file you already keep for continuing competence.
The survey findings are set out openly on the LexisNexis page for Scaling up: The AI-dependent lawyer, though the full report itself asks you to register before you can download it, and Artificial Lawyer reported the research on 2 September 2026.
If you want the drafted-unaided rule written for your firm and fitted to the work your juniors do, that is an afternoon: talk it through with us.