Is It Compliant to Use AI in Case Preparation?

No regulator in England and Wales has banned artificial intelligence from case preparation. The harder question, and the one that actually decides whether your use is compliant, is who stays responsible for the result.

A particular kind of hesitation has settled over litigation teams since the middle of 2025. It shows up in the pause before a fee-earner opens a research tool, in the extra caveat added to an internal note, in the compliance partner who finds it simpler to say no than to work out the answer. The hesitation is rarely about whether AI is useful. Most practitioners already accept that it is. It is about whether using it is allowed.

That is a fair thing to want certainty on, and the honest answer is more encouraging than the nervousness would suggest. No regulator of the legal profession in England and Wales has prohibited artificial intelligence in case preparation. Not the Solicitors Regulation Authority, not the Bar Council, not the Bar Standards Board, not the judiciary. What they have done instead is more demanding and, once you sit with it, a good deal more sensible. They have insisted that the professional using the tool stays competent, accountable and in control of what comes out of it. Get that part right and compliance largely takes care of itself.

What the regulators actually say

The pattern across every relevant body is strikingly consistent. Instead of reaching for a prohibition, each has taken the professional duties that already exist and confirmed that they apply to AI in full.

The SRA has written no AI-specific rulebook, and it has been open about not intending to. Its position, restated in the compliance tips it last updated in February 2026, is that the existing Standards and Regulations already cover the use of AI. A solicitor remains personally responsible for every piece of work, has to verify what a tool produces, must protect client confidentiality, and should make it clear to clients where they are dealing with AI rather than a person.

The Bar reached the same conclusion by its own route. When the Bar Council refreshed its guidance on generative AI in November 2025, it found nothing inherently improper in using reliable AI tools to support legal services, so long as the practitioner understands the technology and uses it responsibly. The Bar Standards Board then published its own guidance in 2026, setting out how the duties already written into its Handbook apply and recommending a risk-based approach before any tool is adopted.

The judiciary looks at the same problem from the bench and lands in the same spot. Its guidance for judicial office holders, updated in October 2025, is permissive but careful. Judges are not barred from using AI, but they hold full personal responsibility for anything produced in their name. It is telling that this guidance has already been rewritten three times since late 2023, a rhythm that says less about indecision than about how quickly the ground is moving.

Read side by side, these are not four different answers to the compliance question. They are one answer, given four times. AI is permitted. Responsibility is not something you can hand off.

The line the courts have drawn

If the guidance sets the principle, the cases show exactly where the line falls, and they are worth knowing in detail, because they are now the cautionary tales every opposing counsel has ready.

The reference point is Ayinde v London Borough of Haringey, heard together with Al-Haroun v Qatar National Bank and decided by the Divisional Court in June 2025. In the Ayinde matter, submissions in a judicial review relied on five authorities that did not exist, one of them presented as a Court of Appeal decision, alongside a misstatement of section 188(3) of the Housing Act 1996. In Al-Haroun, a damages claim valued at £89.4 million, eighteen of the forty-five cases cited proved to be fabricated. In both, the fingerprints of unchecked generative AI were all over the papers.

What the court did with this is the part that matters for anyone weighing up compliance. It did not hold that AI had been used improperly on the theory that AI is itself improper. It held that the lawyers had failed a duty that has nothing to do with technology and everything to do with the job: the duty to be sure that what you put before a court is real. Wasted costs orders followed, along with referrals to the regulators. The warning, aimed at the whole profession, was that responsibility for a submission sits with the person who signs it and cannot be transferred to a machine.

That distinction cuts in favour of careful AI use rather than against it. The failure in these cases was not research assisted by software. It was research that was never checked. The lawyers who came unstuck were the ones who treated a tool's output as a finished product instead of a first draft. Every regulator has since said, in its own words, that this is the mistake to avoid.

Why a preparation aid is the compliant posture

It helps to be precise about what the regulators are anxious about, and what they are not.

They are not anxious about a tool that helps a solicitor find relevant judgments faster, or that shows how a particular court has handled a line of authority, or that flags a shift in the legal climate around an issue before a hearing. They are anxious about a tool that hands over a finished answer that a professional then passes to a court or a client without checking it. The whole distinction is between something that informs a professional's judgment and something that quietly replaces it.

This is why the framing that carries the most weight is preparation aid, not decision-maker. A tool built as preparation does the work the guidance rewards. It speeds up the groundwork, it keeps the professional in the loop at every step, and it leaves every output to be verified against the primary record before it goes anywhere near a pleading. A tool that presents itself, even by implication, as an oracle invites precisely the failure the courts have punished.

The Civil Justice Council put this almost beyond argument in February 2026. Consulting on whether fresh rules were needed for AI in court documents, it proposed no new rules at all for skeleton arguments, statements of case and other advocacy documents, on a single condition: that each bears the name of the legal representative taking professional responsibility for it. The existing duty to verify was judged enough. Offered the chance to build a fence around AI, the system chose instead to reinforce the principle of personal accountability. For anyone designing or choosing a tool, that is the clearest signal available about what compliant actually means.

What compliant AI use looks like in practice

For a litigation team, all of this resolves into a handful of habits that are neither onerous nor unfamiliar.

The first is verification. Every authority a tool surfaces is checked against the primary source before it is relied on, and the primary source, not the summary, is what goes into the papers. The second is retained judgment. The tool prepares; the professional decides, and the professional's name is on the result. The third is transparency, telling clients where AI has been used in the handling of their matter, in the spirit the SRA has described. The fourth is confidentiality, which means understanding where a tool sends data and satisfying yourself that client information is handled lawfully. None of these is new. Each is simply an existing duty, applied to a new instrument.

There is a useful sense of scale to keep in mind here. SRA-commissioned research reported in 2026 that roughly a third of the public had already used generative AI to help identify a legal issue. Clients are arriving with AI in hand whether or not their advisers have adopted it. The professional advantage no longer lies in avoiding the technology. It lies in using it better, and more accountably, than an unassisted layperson can.

The opportunity inside the obligation

There is a reading of all this that treats compliance as a brake on progress. The more accurate reading is that the regulators have described, almost to specification, the kind of AI tool the profession can actually put to work. It does the heavy preparatory lifting, it shows its sources, and it hands the professional something to verify rather than something to trust on faith.

That is the posture Probatur is built around. It surfaces how judges have reasoned, how a line of authority is holding up as later courts revisit it, and how the wider legal climate is shifting on a given question, and it presents every part of that as preparation for a professional's own judgment, with the primary record always a click away. It is designed for the compliant posture the guidance rewards, not in spite of the rules but because of them.

The firms that get the most from this technology over the next few years will not be the ones still asking whether they are allowed to use it. They will be the ones that understood what allowed actually requires, and then chose tools that make meeting that requirement the path of least resistance rather than a daily act of vigilance.

The anxiety, in the end, is pointed at the wrong question. The real question was never whether it is compliant to use AI in case preparation. It was whether you are still the one answerable for the result. Keep the answer to that second question a confident yes, and the first question answers itself.