The story is straightforward: Marsh & McLennan is lobbying US regulators to allow banks to treat credit-risk-insured loans differently for capital adequacy purposes, because demand for AI infrastructure hedging has exploded. But here's what matters to UK mid-market firms regulated by the FCA and PRA: this isn't a technical accounting question. It's a warning sign. If banks can't properly model the risk of a loan to an AI company — so they need insurance to make it fit their capital framework — then the insurance product itself is masking a deeper problem. Regulators loosening rules doesn't solve that. It amplifies it.

This is part of a pattern we've watched accelerate since 2024. The industry built AI products (think of how some legal AI vendors rushed to market with tools like Harvey or Luminance) before it built proper risk frameworks. Now we're seeing the same dynamic in lending and insurance. Banks are lending heavily into AI infrastructure. Insurers are writing policies on those loans. But neither party has actually modelled what happens when an AI project fails, a vendor goes under, or the economic case for an AI system evaporates. So they paper over the gap with regulatory requests and insurance mechanics. That works until it doesn't.

Here's Trovix's view: the answer is not to let banks hold less capital. The answer is to give them better visibility into what they actually own. When a bank writes an AI infrastructure loan, it needs to understand: what is the creditworthiness of the underlying AI vendor? What happens to the loan if the model underperforms? What are the contractual clawbacks? How exposed is this to commodity semiconductor prices or open-source alternatives? These are not questions that insurance solves — and they're not questions that regulatory relief solves either. They're questions that require proper AI risk governance, the kind you build through frameworks like PRA SS1/23 and FCA Consumer Duty PS22/9. Tools like Trovix Audit exist precisely because regulators expect firms to document and audit their AI decisions. That same principle applies to lending and credit risk. You can't insure your way out of not understanding the risk.

What should you do? If you're a mid-market law firm advising on AI infrastructure deals, an insurer pricing policies on these loans, or a financial services firm lending into this space, do not wait for regulatory changes. Build your own model of what AI vendor risk actually looks like. Understand the difference between insurable risk and fundamental business risk. Demand that your counterparties — banks, insurers, borrowers — can articulate their assumptions in writing. And if they can't, that's a signal that the deal is too new and the risk too opaque. Marsh's argument amounts to: 'let us pretend we understand this better by writing it off.' The better argument is: 'let us actually understand it before we write anything off.'

Source: Bloomberg News