Fitch Ratings ran an AI stress test across its rated universe and gave every insurance sub-sector the same business disruption score: 20. On Fitch’s scale, that reads as minor credit pressure and no downgrades expected. The headline writes itself — AI won’t disrupt insurance.
Read Fitch’s own commentary, though, and the exposure it actually flags sits somewhere the headline doesn’t look: inside reinsurers’ models. Not distribution, not apps, not a tech firm stealing the customer relationship. The risk Fitch names is that the correlations between risks shift, the models built on historical data don’t see it, and the mispricing lands in reserves and accumulated exposures. Brokers won’t read that in a rating action. They’ll meet it at renewal, in capacity and price.
What Fitch actually measured
The stress test assessed credit exposure across four sectors — corporates, financial institutions, infrastructure, and structured finance — over a five-year horizon, under three adverse scenarios: over-investment, asset impairment, and business disruption. The insurance score of 20 is the business disruption number.
Scores run from 0 to 100 in 20-point increments, and higher is worse: the score estimates the potential negative impact on an average issuer’s credit rating under the adverse scenario. A 20 means minor credit pressure, with no AI-driven downgrades anticipated over the five years. A 40 signals pressure sufficient to warrant a negative outlook and the possibility of an eventual one-notch downgrade. A 60 means one to two downgrades are likely.
For calibration: retail banks, private banks and wealth managers, and standalone wealth managers all scored 40 — twice insurance’s number. Corporate and investment banks scored 20, the same as insurance. So the 20 is genuinely low relative to the rest of financial services. It is also a specific claim, about rated issuers’ credit profiles, and it should not be stretched past that.
Why insurance scored a 20
Fitch’s stated protections are structural. Regulatory requirements and capital requirements raise the cost of entry. Established distribution networks are already in place. Insurers hold proprietary policyholder data that an entrant cannot replicate by building a better interface. Complex products resist the standardization that makes a market easy to attack.
In other words, Fitch’s case is that a technology firm can build the front end and still not have a license, a capital base, or decades of claims and policyholder history. That is a defensible read of where the moat sits, and for commercial lines especially, it matches what brokers see daily: the product is not the paperwork, it’s the risk judgement and the balance sheet behind it.
The retail exception
Fitch is explicit that the all-clear is not uniform. Retail insurers face greater disruption risk than commercial insurers, because retail products are generally more standardised and consumer preferences can change more quickly. AI could increase competition in retail lines as technology firms develop new ways to reach customers and deliver services, and Fitch says retail carriers may need to adapt their distribution models and customer service to keep pace.
For a commercial broker, this cuts in your favor — complexity is the protection, and commercial placements are where complexity lives. But it is worth noticing that Fitch’s 20 is an average across sub-sectors with different exposures averaged into one number. The average is calm. Parts of the distribution are not.
The fine print: reinsurers and the correlation problem
Here is the passage that matters for brokers, and it is about reinsurers, not primary carriers.
Fitch says reinsurers are less exposed than primary insurers to the risk of AI reducing the role of intermediaries in distribution. Their exposure sits in their core function: modelling and pricing risks. New and changing correlations between risks — particularly in lines of business sensitive to accumulated losses — may be difficult to identify using historical data and traditional modelling methods. The consequences Fitch names are inaccurate pricing, weaker reserving, and the accumulation of unwanted exposures, potentially affecting reinsurers’ financial performance.
In plain terms: a model is a bet that the future resembles the data it was built on, including how risks move relative to each other. If AI changes those relationships — the same systems, models, and vendors spreading across the market at once — a portfolio can become concentrated in ways its own pricing never charged for. This is the same accumulation problem that runs through the current debate about silent AI in primary books; Fitch’s version of it sits one level up, in the reinsurance layer that ultimately prices the capacity brokers place.
That is why this is broker-relevant even though no rating moves. Reinsurance pricing and appetite flow downhill. If reinsurers are mispricing correlated risk, the correction does not arrive as a Fitch press release — it arrives as a capacity pullback, a repricing, or harder terms on the placements a broker is working, discovered at renewal rather than announced in advance.
What a broker does with a 20
Treat the score as what it is: a credit opinion that says the carriers behind your paper are, in Fitch’s view, financially steady against AI disruption for five years. Then keep asking the questions the score doesn’t answer. Four worth putting to carrier and reinsurance-facing partners:
- How are your pricing models validated against correlations that differ from the historical data they were built on?
- How are you monitoring accumulation in lines where AI adoption is concentrated — the same models, vendors, and insureds appearing across the book?
- Where exactly is AI inside the pricing being quoted today — underwriting, triage, claims — and who checks its output?
- When a model’s assumptions break, how quickly does that reach renewal terms?
None of this contradicts the 20. Brokers are already working placements where the AI limit doesn’t exist and coverage has to be structured around a gap the market hasn’t priced yet. Fitch’s score says the institutions are stable while that gets worked out. It says nothing about whether any individual risk is priced right — which is the part of the market a broker actually trades in.
The honest limits
Three cautions before citing the 20 at a client.
First, it measures credit-rating impact on an average issuer under an adverse scenario. It is not a measure of broker economics, commission pressure, or service competition, and Fitch does not claim otherwise.
Second, the horizon is five years, and Fitch’s scenario design is explicit about that window. The stress test does not address what the picture looks like in year six, when the structural protections it credits — data, distribution, regulation — have had longer to erode or entrench.
Third, the 20 is about disruption, not adoption. Insurers are already using AI to grow: in an Accenture survey of 263 senior insurance executives, reported by Insurance Asia, 81% said their data and AI initiatives had increased gross written premiums by at least 5%. AI as a growth tool inside incumbent carriers and AI as a disruptive force against them are two different claims. Fitch scored the second one. The first one is already running.
The useful summary of the 20 for a broker: stable carriers, a live modelling question in the reinsurance layer, and a retail exception to watch. That is a calmer picture than the disruption headlines — and a more specific one than the all-clear headline suggests.
Sources: Fitch Ratings’ AI stress test, as reported by Insurance Asia, “Insurers face AI competition as technology firms target retail” (Nov 2, 2026); Accenture survey figures via the same report.
This article is for general information only and is not coverage, legal, or investment advice. Placement and coverage decisions should be made with your own broker and counsel on the specific facts of the risk.