Marsh & McLennan wants US regulators to let banks hold less capital against loans that are covered by credit risk insurance. This matters urgently to UK legal, insurance and financial services firms because it reveals a fault line in how the industry is financing AI infrastructure. UK banks and their PRA regulators face the same problem: demand for AI data centre lending has exploded, but the traditional tools—including credit insurance—don't satisfy either banks' risk appetites or the Prudential Regulation Authority's requirements under PRA SS1/23. Regulators built capital frameworks around predictable asset classes. AI infrastructure lending is neither predictable nor mature. Loosening the rules won't fix that.

What Marsh is really saying is this: the insurance industry can't price AI loan risk accurately, so banks are trying to shift that problem onto the capital regulation framework instead. This is a pattern we're seeing across regulated firms globally. When facing genuine uncertainty—whether it's in legal discovery automation (where tools like Luminance promise to cut review costs but still need human lawyers to validate findings), or in claims triage (where generalist AI chatbots replace intake forms but create compliance gaps under FCA Consumer Duty PS22/9), firms are tempted to ask: can we loosen the rules rather than build better governance? The answer is always no. Loose rules plus unclear risk equals systemic risk.

Trovix's view is blunt: this story shows why AI governance cannot be delegated to product vendors or regulatory arbitrage. Marsh is effectively saying 'we can't model this risk properly, so please relax oversight.' The FCA and PRA should hear a different signal. UK firms betting on AI loan portfolios need Trovix Audit governance—a real-time map of where AI systems are making material credit or insurance decisions, who trained them, what data they used, and how they fail. This is not theoretical. If you are a mid-market insurance broker, bank, or accountancy firm writing AI into your underwriting or advisory workflow, you cannot rely on off-the-shelf tools from Microsoft Copilot or generic enterprise AI platforms. Those tools don't know your regulatory obligations. They don't log decisions in ways that satisfy FRC ISA UK or SRA Code auditing. They don't warn you when they're handling data outside their training distribution—which in AI lending, happens constantly.

Here is what to do now. First: audit which of your client-facing or internal processes have already deployed general-purpose AI (including LLMs). Second: map those processes against your specific regulatory requirements—PRA SS1/23 if you're regulated for capital, FCA Consumer Duty PS22/9 if you touch retail, EU AI Act obligations if you process EU data, ICO UK GDPR if you're handling personal data. Third: implement Trovix Watch to track regulatory moves on AI in your sector in real time, because capital rules, insurance rules, and AI governance rules are all moving simultaneously and regulators are not coordinating them cleanly. Fourth: do not wait for your regulator to demand governance—build it now, show evidence of it, and use it to negotiate with your board and your auditors. Marsh's lobby for easier capital rules reveals that the industry is still hoping regulation will soften around AI. That hope is misplaced. The rules will tighten. What tightens faster is enforcement.

Source: Bloomberg News