Your policy was written before your business used machine learning. The gap it left is specific, predictable, and already costing companies like yours.
The real fear isn’t that your insurance is wrong. It’s that your insurance looks right.
The renewal went through. The certificate is on file. Your broker confirmed cover. Everything appears exactly as it should.
But between the date your policy was last properly worded and the date your business started using AI in its products, its processes, or its decisions, a gap opened. Nobody flagged it. Not your broker, not your insurer, and certainly not you. Because from the outside, nothing changed.
From the inside, everything did. And in the insurance market, this AI insurance gap already has a name.
The Gap Opens Quietly
Underwriters call it “drift.” The business moves; the policy doesn’t.
It rarely starts with a dramatic decision. Nobody calls an all-hands meeting to announce “we’re now an AI company.” Instead, a team adopts a new tool. A workflow gets automated. A product feature starts using machine learning to personalise recommendations. None of it feels like something you’d mention to your broker. It feels like an upgrade.
But your policy was rated, priced, and worded based on what your business did before those changes. The underwriter assessed a company that no longer exists.
The insurance industry has seen this before. Between 2015 and 2020, insurers discovered they were paying cyber-related claims under traditional property and liability policies that never mentioned cyber risks. Lloyd’s estimated the exposure ran to hundreds of billions. In 2019, Lloyd’s mandated that all policies provide clarity on cyber coverage. Kennedys Law now warns that “silent AI” presents exactly the same pattern: risks neither explicitly included nor excluded, sitting unnoticed in existing policy wordings until a claim forces the question.
The gap between your declared risk profile and your actual risk profile is where claims get declined. And that gap is widening with every AI tool your team adopts.
Where It Breaks
Three specific points. Each one is a door you can check today.
Professional indemnity. Your PI cover was worded around human professional judgement. If AI now contributes to or shapes your client deliverables, the basis of the cover may not match the basis of your work. Kennedys and the Forum of Insurance Lawyers have identified this directly: traditional PI policies were designed to cover errors resulting from human actions, and as AI becomes integral to decision making, standard wordings may not respond to claims arising from AI-generated outputs.
Product liability. If your product learns or adapts after deployment, the liability profile at point of sale is different from the liability profile six months later. Your insurer priced a static product. You’re now selling a dynamic one. Most product liability wordings weren’t built for that.
Cyber. Your cyber policy likely covers data breach and system failure. But does it cover an AI model being manipulated, producing harmful outputs, or making decisions based on poisoned training data? Swiss Re’s September 2024 analysis found that AI risks are neither explicitly mentioned, limited, nor excluded in most existing policy language. For most policies written more than eighteen months ago, these exposures sit in silence.
https://www.techlifefuture.com/ai-insurance-exclusions-sme
And the market is responding. From January 2026, Verisk introduced new general liability exclusion forms that allow insurers to carve out generative AI exposures entirely. Major insurers including AIG and WR Berkley have filed to exclude AI-related liabilities from commercial policies. The silent coverage you may have relied on is being removed, often without fanfare, at your next renewal.
And Underneath All of It: Disclosure
The Insurance Act 2015 requires you to make a fair presentation of your risk. If your business has materially changed how it operates by adopting AI and you haven’t told your insurer, you may have breached that duty.
This isn’t theoretical. In Berkshire Assets v AXA [2021], the High Court held that failure to disclose material circumstances at renewal entitled the insurer to avoid the policy entirely. The principle is settled: material change plus non-disclosure equals remedies that range from adjusted terms to full avoidance.
There is no AI-specific UK case law yet. But the legal framework is tested, the market is acting, and the first contested AI claim will set the precedent. The question is whether your business is on the right side of disclosure when that happens.
The Gap Doesn’t Announce Itself
The AI insurance gap won’t appear on your certificate of insurance. It won’t trigger a warning from your broker. It will only become visible when you make a claim and your insurer asks: “Was this the business we agreed to insure?”
The founders who check now, before that question is asked, are the ones whose cover actually works when it matters.
