Anthropic's expanded Claude for Legal — launched with integrations into Westlaw, DocuSign and Box — is genuinely capable. Deposition prep, case research, document drafting, privacy analysis: these are high-value tasks that consume partner and associate time. For mid-market UK law firms, the appeal is obvious: plug it in, reduce billing-critical overhead, move faster. But this is exactly the wrong way to think about it. A tool that generates legal documents in seconds is only as good as the firm's ability to explain, audit and justify how it was generated — to clients, to the SRA, and to itself. Anthropic has built a powerful engine. It has not built the governance layer that UK regulated firms actually need.

This story is part of a broader pattern. Harvey, Legora, Luminance, Microsoft Copilot — every credible legal AI vendor is now racing to offer point-solution features: search faster, draft faster, research faster. They are all shipping capability and assuming governance will follow. It won't. The legal tech market has learned nothing from the cloud adoption curve of 2015–2018, when firms rushed to implement SaaS tools without understanding data residency, access controls or audit trails. The EU AI Act now treats high-risk AI in legal services as a compliance obligation, not a competitive feature. The SRA's Technology Enabled Business programme and the ICO's emerging guidance on generative AI in legal work mean UK firms face real regulatory friction when they cannot demonstrate control. A shiny tool without provenance and auditability is a liability dressed as an efficiency gain.

Trovix's view is this: the bottleneck in legal AI is not capability. It is confidence. A mid-market firm already knows it needs faster document analysis and smarter research. What it does not have is a way to operationalise that knowledge safely. That is why we built Trovix Audit as a governance layer, not a feature layer. When you integrate Claude — or any LLM — into your workflows, you need to see which documents were analysed, which prompts were used, which outputs were checked and by whom, and whether the decision was logged in a way that satisfies the SRA Code. Luminance has built this alongside model capability. Harvey has started to. Anthropic has not. That gap matters most to the firms that need AI most urgently: the ones operating at the scale where quality control is manual and margin is tight.

If you work at a mid-market law firm, insurer, accountancy practice or financial services firm considering Claude for Legal or any similar tool, do this: before you integrate it, ask the vendor three questions. First: who owns the data I send into this system, and where does it live? Second: can I see and audit every decision the AI made, including what it considered and rejected? Third: if a client asks me to explain how a document was drafted, can I produce a complete audit trail? If the answer to any of these is 'that is not our responsibility' or 'the system does not track that', you are not ready to deploy it. Not yet. Build the governance infrastructure first — through tools like Trovix Audit and Trovix Sift — and only then integrate the capability layer. Speed without control is just risk with a better interface.

Source: TechCrunch