Buying a tool and writing a policy achieves little if the team understands neither. Training is the step firms skip, and it is the one that decides whether AI helps or harms.
Teach the why, not only the how
People follow rules they understand. Explain why client data cannot go into a free chatbot, why a fabricated case is a serious matter, and what the firm's duties are. The how follows easily once the why has landed.
Three duties carry almost everything you need to teach. Confidentiality decides what a fee earner is allowed to type into a prompt. Competence decides whether anyone in the room can tell confident output from correct output. Supervision decides who answers for the result once it leaves the building. Tie each rule back to one of those and the session stops sounding like an IT briefing and starts sounding like the work.
The SRA warning notice on the misuse of AI, published on 17 August 2026, is blunt on the confidentiality half of this. Paid products as much as free ones store what you type, keep it longer than you assume, or use it to train a model, and the notice expects contractual, technical and organisational safeguards in place before client material enters any tool.
Start with the failure modes
The mistakes these tools make are predictable, which makes them teachable in an hour. A general model generates text by predicting what tends to follow what. It holds no index of real cases and no sense of truth, so it produces something shaped exactly like a citation with nothing real behind it, and the confidence of the output tells you nothing about its accuracy.
Give the point some weight by showing what it has cost. In R (Ayinde) v London Borough of Haringey, decided in 2025, a Divisional Court led by the President of the King's Bench Division dealt with submissions citing authorities that did not exist. It referred the lawyers responsible to the Bar Standards Board and the Solicitors Regulation Authority. An AI hallucination case tracker recorded about 1,598 cases worldwide as at 9 June 2026, fifty-nine of them in the United Kingdom. Put those figures in front of a room once and the checking rule needs no further defence from you. The habit that follows from them is set out in verify before you file.
Keep it short and real
A long seminar is forgotten by Friday. Short sessions built around the firm's own work, with real examples of good and bad use, stay with people. Cover the approved tools, what they are for, and the checks expected before anything goes out.
Use work that passed through your office rather than a worked example from a vendor. Pick three pieces of recent work, show where a tool helps and where it goes wrong on each, and let the room argue about the difference. Our own half-day literacy session is built that way for up to twenty people at a time.
Show how the learning was identified
Training that leaves no trace is hard to rely on when somebody asks. The SRA reviewed 123 training records from criminal and civil practitioners across 66 firms for its fourth annual assessment of continuing competence, published on 20 August 2026. Almost every record existed and was current. Almost half failed to explain how the learning need was identified and why it mattered, which is the part that turns a list of attended sessions into evidence of reflection.
The regulator now assesses whether a solicitor is aware of and uses warning notices, guidance and the Statement of Solicitor Competence when planning their learning, and the AI warning notice arrived three days before that assessment was published. So for each person using a tool, write down what changed in their work, what that exposed as a gap, what they did about it and whether it worked. The full exercise, including what the SRA has consulted on for the 2027/28 practising year, sits in continuing competence and AI.
Train your juniors differently
LexisNexis surveyed 543 UK legal professionals for a report published on 2 September 2026. Three quarters of them think junior lawyers who use AI heavily will struggle to develop legal judgement, and one in ten said they are dependent on the technology to do their jobs. Keeping trainees away from the tools answers none of that, since it trains them for a job that will not exist.
Set the order of work instead. On the tasks that teach most, a first piece of research, a first advice, a first witness statement, the junior drafts unaided and runs the tool over the finished draft to compare the two. 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.
Check whether the training took
Attendance proves nothing, and a firm that counts licences and calls it adoption has measured the wrong thing. Pick one task the training was meant to change, write down how it works today and roughly how long it takes, then count three things over ninety days. Count how many matters used the tool on that task, the time from instruction to a usable first draft, and how often the supervising fee earner sent that draft back. The third number tells you whether speed came at the cost of quality, which is the one a partner most needs before a renewal. At ninety days you widen it, fix the training or drop it at renewal.
Make it ongoing
The tools change, and so do the risks, so a single induction is not enough. Refresh the training, and fold AI into how new staff are brought on.
Supervisors need a version of this aimed at them. The warning notice closes a door some firms have been leaning on, because the supervisor of a junior who used a tool answers for the output as fully as the person who wrote the prompt. Tell them what to look for in AI-assisted work, which is confident wording that misstates the law, citations that do not exist and summaries that miss the point. A polished draft is not a checked one.
Confident, well-trained people are the firm's best protection. Train them poorly and you spend the year correcting mistakes. Train them well and keep at it, and the tools earn their place.
The SRA's compliance tips on AI and technology make a good handout for the first session.
Our literacy sessions do exactly this in half a day, for up to twenty of your people, with a workbook each and a thirty-day action card: see what they cover.
