AIG's warning that AI data centre buildout is pushing P&C providers to their limits should land hard with UK regulated insurers and the underwriters who work for them. The message is clear: comprehensive coverage for AI infrastructure — from project finance through operations, including fire suppression failures, power grid dependencies, supply chain disruption, and novel cyber vectors — sits outside the risk appetite and technical knowledge of insurers built for traditional manufacturing and real estate. This is not AIG being cautious. This is AIG saying the industry's underwriting models, loss history, and actuarial frameworks do not yet exist for this asset class. For UK insurers bound by PRA SS1/23 governance expectations and FCA Consumer Duty obligations, that admission should trigger immediate board-level questions: what do your underwriters actually understand about AI infrastructure risk, and what are you underwriting blind?

This is part of a wider pattern. For two years, the insurance industry has treated AI as a tool — bringing in vendors promising automation, document review, claims triage, and faster underwriting through large language models. Systems like Harvey for legal tasks and basic claims routing bots promised efficiency. What they actually delivered was speed without insight. Insurers adopted AI chatbots for customer service and document automation for claims without building the governance frameworks, bias testing, or audit trails that PRA SS1/23 and ICO UK GDPR demand. The result: regulatory risk on top of market risk. Now, as actual AI infrastructure — the chips, the power, the data centers — becomes the thing being insured, the industry is discovering it never built the competency to price it. The problem is not that insurers lack capacity. The problem is that they lack understanding.

Here is Trovix's view: most of the AI products deployed in insurance over the past 18 months have been answers to the wrong question. They have asked 'how do we make underwriting faster?' instead of 'how do we understand what we are underwriting?' That distinction matters. When you push document automation and generative AI into claims triage without first mapping what your firm actually knows about a claim, what your firm is liable for, and where your AI is making decisions that could breach FCA Consumer Duty or SRA Code principles, you are not accelerating — you are documenting negligence. The firms winning with AI are not the ones running the fastest chatbots. They are the ones building structured governance: knowing what data flows into underwriting decisions, testing outputs against historical loss data, maintaining audit trails, and being able to explain to a regulator why a claim was approved or denied. This requires different tools entirely — not faster AI, but visible AI. Not inference engines, but oversight infrastructure.

If you are a mid-market insurer or underwriter right now, the practical move is not to chase the capacity shortage by buying more AI tools. It is to audit what you have already deployed and honestly map what your firm understands about the risks you are pricing. Ask: can your underwriters explain in plain language why your AI system approved or denied a claim? Can your compliance team produce an audit trail that satisfies the FCA's principles on algorithmic decision-making? Can your actuaries actually test AI-assisted underwriting decisions against real loss patterns, or are you guessing? If the answer to any of those is 'not yet', then you need Trovix Audit before you need faster automation. Governance is not the cost of doing AI. It is the only way to do it profitably.

Source: Bloomberg News