Forbes reports that £2.1 billion has flowed into legal-tech startups globally in H1 2026, with models like Soxton layering licensed attorney review over AI-generated documents to cut costs for small businesses. This matters urgently to UK mid-market firms because the story reveals the precise problem they now face: a flood of cheap, unregulated AI tools that promise to commoditise routine work. The SRA Code of Conduct 2019 is clear — you are responsible for supervision and quality. If you deploy an AI tool that produces defective documents, the liability falls on you, not the vendor. The FCA Consumer Duty (PS22/9) applies equally to authorised firms holding client money or delivering advice. Yet most legal-tech startups operate without audit trail, explainability, or segregation between AI generation and human review that the SRA would actually accept.

What this £2bn represents is a market correction around commodifiable work — due diligence data extraction, contract term indexing, precedent templates, intake triaging. That work should move to AI. But the funding trajectory tells us something darker: venture capital is backing speed and volume over governance. Companies like Harvey and Legora have raised eye-watering sums partly because they marketed generative AI as a replacement for knowledge workers, not as a tool for them. The market has since corrected. What's emerged is a two-tier industry: products built for unregulated bolt-on use (document generators, solo practitioners, small shops) and products built for regulated deployment (audit trails, segregated workflows, explainability logs, version control). Most of the £2bn is funding the first type. UK regulated firms are vulnerable to adopting those tools and then discovering, mid-matter, that they cannot prove compliance with the SRA's Technical and Operational Standards for AI in legal services.

Trovix's view is this: if you are a regulated firm, cost savings are not your primary criterion. Compliance architecture is. The vendors getting the most attention — and the most funding — are not necessarily building for the constraints that actually bind you. A tool like Microsoft Copilot in legal, or a generic document-generation API, is cheaper upfront because it does not embed supervision, audit, or the ability to log why an AI decision was made or rejected. Luminance and some of the Harvey use-cases do better, but they remain largely designed for large firms with dedicated AI governance teams. Mid-market practices need something different: AI that works *within* your existing quality gates, not around them. That means document intelligence that surfaces uncertainty and flags risk, not just extraction. It means RAG-based assistants that cite their source and degrade gracefully when out-of-domain. It means intake automation that learns from your closed matters and leaves a traceable record.

Here is what you should do on Monday: audit which AI tools are already in use in your firm. For each one, write down (a) whether the vendor has liability insurance for AI-generated output, (b) whether they hold a copy of every output for audit purposes, (c) whether you can explain to the SRA exactly how a junior was prevented from sending a defective document. If you cannot answer all three, that tool is not fit for a regulated environment, regardless of cost or time-saving. Then talk to your practice management team about which workflow problems actually need solving. Is it document review? Use Trovix Sift. Is it matter intake that creates clean data for downstream systems? Use Trovix Brief. Is it keeping associates up-to-date on regulatory change? Use Trovix Watch. The £2bn boom is real, but it is not your market. Your market is compliance-first, cost-second, built on audit and explainability.

Source: Forbes