On 23 October, Dext launched AI Connector—a tool that lets accounting and bookkeeping practices query their operational and client data using Claude, ChatGPT, and Copilot. The pitch is simple and seductive: no extra cost, natural language access to practice intelligence. For UK mid-market accountancy firms navigating the FCA Consumer Duty (PS22/9) and increasing ICO scrutiny of third-party data handling, this announcement should trigger a specific question: who owns the governance risk when a fee-earner uses ChatGPT to pull sensitive client data into an uncontrolled conversation thread? Dext has built a retrieval layer. It has not built a governance framework.

This is part of a wider pattern in regulated AI: vendors are racing to offer conversational access to sensitive data because it feels intuitive and immediate. Harvey does this for legal workflows. Luminance does it for due diligence documents. Microsoft Copilot does it for enterprise data. The assumption is that if you can ask a question in plain English and get an answer, the problem is solved. It is not. The FRC's emerging guidance on AI governance in audit (building on ISA UK 240) makes clear that firms remain liable for outputs they act upon, regardless of how easy the tool makes generation feel. The ICO's recent positions on GDPR and algorithmic decision-making reinforce that convenience and compliance are not the same thing.

Trovix's view is blunt: retrieval without safeguards is a liability vector dressed as innovation. Yes, practitioners want fast access to data. But 'fast' and 'governed' are not opposites. A responsible AI integration architecture should enforce four things Dext's model does not: (1) explicit audit trails of every query and result returned; (2) role-based access controls tied to practitioner authority and client engagement scope; (3) mandatory review gates before sensitive data moves outside the firm's systems; (4) firm-controlled output handling, not consumer LLM terms of service. Trovix Aria and Trovix Watch sit inside this kind of framework. They give practitioners the conversational experience they want while keeping data, decisions, and accountability where regulators expect to find them—inside the firm.

For a mid-market accountancy practice right now, the move is not to rush into Dext's AI Connector. It is to ask yourself: does your firm have documented AI governance policies aligned with your professional body's standards and your client engagement letters? Have you classified which data can safely enter a third-party LLM and under what conditions? Have you trained your team on the difference between an AI tool that enhances decision-making and one that outsources it? If the answer to any of these is 'no', then free or cheap access to your data through ChatGPT is not a feature—it is a compliance risk waiting to crystallise. Invest in the governance framework first. The tools will follow.

Source: IT Brief Australia