The New Statesman's April investigation found that LLM-generated text has entered UK acts of parliament and that AI analysed departmental spending bids in Rachel Reeves' June 2025 Spending Review. Meanwhile, the government awarded £476 million in AI consultancy contracts, with 60 House of Lords members holding declarable interests in AI companies. For mid-market legal, insurance, financial services and accountancy firms, this matters urgently: if Whitehall cannot establish basic audit trails and transparency around its own AI use—while managing billions in public money and regulatory frameworks that govern your clients—then the regulatory framework for your AI deployment is being written by organisations that do not yet understand what they are doing. The FCA Consumer Duty PS22/9, SRA Code, PRA SS1/23 and ICO UK GDPR all now require documented AI governance. Government departments appear to be ignoring this principle at scale.

This story is part of a pattern. Generic LLM tools like ChatGPT and Microsoft Copilot are being deployed into high-stakes contexts—legislation, budget analysis, procurement decisions—without the governance infrastructure that regulated firms are expected to have. The EU AI Act's classification of government AI as high-risk is not theoretical; it is already happening in the UK. What the New Statesman reveals is that when organisations adopt AI rapidly, they treat governance as optional. They do not build explainability into procurement decisions. They do not maintain audit trails. They do not track who made the final decision or what the AI actually suggested. And when those organisations are the ones writing the rules for your compliance obligations, you inherit the chaos. Firms using Luminance or Harvey for document review have at least invested in purpose-built legal AI; firms rolling out Copilot across legislation drafting have not.

Trovix's view is blunt: the government's AI use—and the absence of governance around it—proves that generic LLMs without audit infrastructure should never be trusted with high-stakes regulatory or financial decisions. The story does not prove AI itself is the problem; it proves that AI without governance is. A regulated firm deploying AI into underwriting, client advice, tax calculation or compliance flagging needs more than a powerful model. It needs a documented chain of reasoning, version control, human sign-off audit trails and the ability to say exactly what the AI saw and what it recommended. This is not what Copilot or ChatGPT provide by default. It is what Trovix Audit is designed to deliver: a compliance and governance dashboard that makes AI decisions visible, traceable and defensible to regulators. The government's failure is instructive precisely because it shows what happens without this layer.

If your firm is still treating AI implementation as a technology decision rather than a governance decision, today is the day to recalibrate. Audit your current AI use: what tools are in production, what decisions do they support, who verified the outputs, what traces exist? The FCA, SRA, PRA and ICO will expect answers. Appoint someone to own AI governance formally—not the CTO, the compliance officer. If you are deploying Copilot or similar tools, add a mandatory human review gate and document every decision. Consider Trovix Watch for continuous monitoring of regulatory changes in AI governance; this landscape is moving monthly now. Most urgently: do not assume the government knows what it is doing. It clearly does not. That is your competitive advantage and your legal obligation.

Source: New Statesman