Forbes reports that £2.1 billion flooded into legal-tech startups in the first half of 2026, with venture capital backing products like Soxton that pair AI-generated documents with licensed attorney review. The narrative is seductive: automation makes legal services cheaper, therefore more accessible to small businesses. But this story misses the actual problem facing mid-market UK regulated firms. The FCA Consumer Duty (PS22/9) and SRA Code of Conduct for Solicitors demand that firms understand, govern and remain accountable for every output they distribute to clients — whether drafted by a human or an LLM. Investment in flashy LLM-plus-review models does nothing to solve the governance barrier that still prevents most firms from safely deploying legal AI at scale.
What we're witnessing is the fourth wave of legal tech hype in a decade. First came contract analytics (Luminance, Legora). Then came legal research on steroids (Harvey). Then came general-purpose LLM wrappers (Microsoft Copilot for legal, Google's new Cloud offerings). Each wave attracted capital and delivered genuine productivity gains to firms with sufficient technical and compliance maturity to implement them properly. But each wave also left behind a graveyard of botched deployments: firms that bought the tool, couldn't integrate it with their existing workflows, failed to document audit trails, and quietly stopped using it after six months. The £2 billion investment in 2026 assumes the problem is access and cost. The real problem is operational readiness and regulatory risk.
Trovix's position is unambiguous: before any UK regulated firm (law firm, accountancy practice, insurer, financial services house) deploys legal or document AI, it needs working answers to three questions that venture-backed startups are not designed to solve. First, do you have the audit capability to prove what the AI did and why? (This is why Trovix Audit exists — the ICO UK GDPR and ISO 42001 now explicitly require AI governance records.) Second, can your fee-earners actually use the AI within their existing case management and knowledge systems without manual workarounds? (This is Trovix Aria — a knowledge assistant that sits inside your tech stack, not alongside it.) Third, is the output actually fit for your specific client work, or are you deploying a generic model that will require 60% rework? Soxton's model of AI-draft-plus-attorney-review is honest about the cost. But if your firm still lacks the infrastructure to track which attorney reviewed what, on which date, for which client, you are still exposed under the SRA Code. The venture capital model succeeds by moving volume; the compliance model succeeds by moving slow and auditable.
If you run a mid-market law firm, accountancy practice or insurance operation, here is what you should do on Monday morning. Stop evaluating AI tools by how many things they claim to automate. Start evaluating them by whether they include or integrate with governance layers that satisfy your existing compliance framework. Ask vendors whether they can report to you in real time: what output came from the AI, which human reviewed it, when, and to what standard? Ask whether the tool plugs into your existing workflows or requires your people to learn new systems. Finally, ask your compliance officer to review the vendor's security certification, data residency policy and indemnity clauses under the light of the EU AI Act as it increasingly shapes UK regulatory thinking. The £2 billion in venture capital is real, but it is not solving the problem your firm actually faces. That problem requires slower, deeper, more deliberate implementation. It requires governance-first thinking. It requires honesty about what AI can do in a regulated environment, and what it cannot.
Source: Forbes