FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Aipedia: Solicitors and law firms
Technology-assisted review is accepted in English disclosure when the method is suitable and proportionate, but the solicitor remains responsible for what is disclosed and withheld. A mis-coded privileged document can harm a client's case and lead to a negligence claim.
Part of: AI in solicitors and law firms
In short
Courts have approved predictive coding since Pyrrho v MWB in 2016, where the High Court found it suitable for a review of over three million documents. Generative AI now adds summarising, issue coding and privilege suggestions. The risks are an indefensible methodology, missed relevant documents and privileged material disclosed by mistake. Code 2.1 to 2.2, 1.4 and 3.5 apply, and confidentiality matters when documents go to vendor platforms. A client's claim for a lost advantage is usually covered under the SRA minimum terms, subject to the policy terms.
Disclosure teams use technology-assisted review (TAR), also called predictive coding, to rank documents by likely relevance after a lawyer trains the system on a sample. Newer generative AI tools summarise documents, suggest issue codes, draft privilege log entries and answer questions across a document set. Both can cut review time and cost dramatically, which is why clients and courts expect them in large cases.
In Pyrrho Investments v MWB Property [2016] EWHC 256 (Ch), Master Matthews approved the parties' agreed use of predictive coding in a review of more than three million documents. The process involved training on a senior lawyer's sample, quality-assurance sampling and repeated rounds of review. The judge found no evidence that it was less accurate than manual review and that it promoted the overriding objective, but said future approval would depend on the circumstances of each case.
That last point matters. A methodology must be explained, agreed where possible and validated. If the other side challenges the review, the firm needs to show what was done, how recall was tested and who made the judgements.
The SRA's warning notice lists the evidence duties in Code 2.1 to 2.2, Code 1.4 (not misleading the court), Code 2.5 (court orders) and Code 3.5 (supervision) among those engaged by AI. Code 6.3 covers confidentiality of client documents uploaded to review platforms, and the notice expects contractual, technical and organisational safeguards. The SRA's supervision guidance requires human review, scrutiny and professional judgement over AI outputs.
Privilege review is where AI errors hurt most. A tool that misses a privileged email, or codes an advice note as merely relevant, can lead to it being disclosed. Courts can sometimes restrict use of documents disclosed by mistake, but that is not guaranteed and the damage may already be done. The Chancellor of the High Court's April 2026 speech noted that confidentiality is a prerequisite for privilege and that public AI tools appear not to be confidential, so where review happens also matters.
Illustrative: in a commercial dispute, a firm uses a generative AI tool to suggest privilege calls across 200,000 documents, with lawyers checking only the documents the tool flagged. Several emails between the client's in-house lawyer and the board are coded as not privileged and disclosed. The other side uses them in cross-examination, and the client's position in settlement weakens. The client later claims against the firm.
A client's claim for a lost advantage or increased settlement is a civil liability claim, which the SRA minimum terms require to be covered, with defence costs in addition. The closed list of permitted exclusions does not include AI. Causation can be contested, because the firm may argue the outcome would have been the same. See a privilege and data breach scenario and preserving privilege when notifying insurers. A vendor platform data breach may also engage a cyber policy.
Apex arranges PI for solicitors and law firms with disputes practices. A broker can review how your policy treats lost-litigation and privilege claims. See the AI in law firms hub, AI, confidentiality and privilege and AI and witness statements.
Yes, where suitable. In Pyrrho v MWB (2016) the High Court approved agreed predictive coding for over three million documents, finding it proportionate and no less accurate than manual review. Approval depends on the circumstances of each case.
Act quickly to seek its return and restrict its use, notify your insurer if a claim is possible, and review the process that failed. Whether use can be restricted depends on the circumstances, so prevention through human privilege review is far better.
Usually, a client's negligence claim is covered civil liability under the SRA minimum terms, which contain no AI exclusion. Causation and loss will often be disputed. Cover is subject to the policy terms, and any top-up layer has its own wording.
Only to a tool with proper safeguards. The SRA notice expects contractual, technical and organisational protections, a secure environment and no training use unless authorised. The Chancellor's 2026 speech suggested public tools appear not to be confidential, which puts privilege at risk.
Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.
Start a PI quote Request a callbackApex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.
