Bloomberg Law's latest survey should stop mid-market UK law firms, accountancies and financial services practices mid-click on their next AI purchase. Three-quarters of the 1,000+ attorneys surveyed are genuinely concerned that AI systems will malfunction or behave unpredictably. A quarter fear job displacement. But here's what should actually concern you: these concerns are not irrational scepticism. They are correct. And the fact that 76% of the profession hasn't yet built the governance frameworks to manage that risk means most AI deployments in UK regulated firms are currently operating blind to the SRA Code, FCA Consumer Duty PS22/9 and the principle-based compliance regimes these firms are supposed to follow. The Bloomberg data is not a problem statement about AI itself. It is a data point about a crisis in implementation culture.

The story sits at a hinge moment. For the past two years, the AI narrative in legal and professional services has been dominated by tool adoption: Harvey, LLMs, Luminance's document analytics, Microsoft Copilot, and dozens of others. Each has genuine utility. The conversation has been: which tool, how much, how fast? What we are now seeing — reflected in this survey, and in parallel moves by the FRC, the Lloyd's Blueprint Two governance standards, and the EU AI Act — is a hard shift towards: does this tool produce harm, who is liable, and can we prove we managed it? The ROI question the Bloomberg headline poses is almost a distraction. The real question is: do you have a documented, auditable AI risk and governance framework that regulators will accept? Most UK firms, frankly, do not.

Here is Trovix's honest view. Building trust in AI inside a regulated firm means three things: first, you need visibility into what your AI systems actually do with client data, advice, and decisions — not aspirational claims from vendors, but real, documented behaviour. Second, you need to map AI usage to specific regulatory obligations and client expectations, not deploy tools because competitors have them. Third, you need a way to prove to the FCA, SRA, FRC or ICO that you have managed the risks documented in surveys like this one. Most AI implementations in the market skip straight to 'go fast'. They skip the governance layer. Tools like Trovix Audit exist precisely because 'trust the vendor' is not a compliance strategy. If your firm's answer to 'why do you use this AI?' is 'it seemed helpful' or 'everyone else does', you are already failing your Consumer Duty obligations and your professional responsibility framework. The best tools in the market — and Harvey and Luminance are serious products — still require this discipline around them. The survey tells us most firms have not yet built it.

So what do you do on Wednesday morning? First: stop treating AI adoption as a single decision. It is not. Treat each use case separately, starting with the ones that touch client advice, data handling or regulated outputs. Map each against your SRA Code, FCA obligations, and your firm's own risk appetite. Second: demand visibility. If your AI vendor cannot explain what their system does with your data, or if they refuse audit rights, they are not enterprise-ready for regulated firms. Third: build a governance layer, which means someone in the firm owns AI risk, someone audits it regularly, and you have a documented trail. This is not bureaucracy for its own sake — it is the price of operating in a regulated sector, and it is also the thing that makes AI actually work. Firms that have done this are getting real ROI. Firms that skipped it are now scrambling.

Source: Bloomberg News / PR Newswire