AI Momentum 98.5 +1.5%GitHub Pulse 5503 +6.5%Capital Flow $204.0B +19039.3%Hiring Heatmap 0.0 -100.0%
Advertisement
P(doom)

Allianz says insurers are badly undercounting AI losses, and one faulty model could hit many industries at once

Allianz Commercial, the corporate insurance arm of Germany's Allianz, says the AI losses insurers can see today are a small share of the real bill. In a report released November 19, it argued that many AI-driven losses are being paid under existing cyber, product liability and crime policies, or under directors-and-officers (D&O) cover, which protects executives when they are sued, without ever being recorded as AI claims.

That matters to any company buying insurance or rolling out AI at scale. When insurers cannot measure a risk, they tend to price it high, exclude it, or both. In the US, exclusions are already spreading.

Allianz Commercial says it is already tracking AI-related claims worth millions of euros. It estimates that two kinds of cases it follows could each cost the insurance industry roughly $100 million in total. One is vehicle accidents allegedly caused by faulty AI safety systems. The other is shareholder lawsuits claiming executives misled investors about AI. Both figures are modeled scenarios, not money paid out.

The losses insurers can and cannot see

The gap between worry and recorded losses is wide. AI jumped from 10th to 2nd in Allianz's 2026 annual survey of business risks, the biggest rise of any category. Yet the company says its confirmed AI claims would not place AI among the top 20 causes of loss for commercial insurers.

Outside counts point the same way. The MIT AI Risk Initiative logged 453 reported AI incidents in 2025, up from 179 in 2023, a rise of more than 150%. Reported incidents are not the same as insured losses. Cornerstone Research counted 16 AI-related US securities class actions in 2025, up from 7 in 2023. The D&O Diary, a legal blog that tracks such suits, counted 22 by the end of August 2026.

Chief executive Thomas Lillelund said the "apparently modest number" of confirmed claims hides companies' true exposure. Michael Daum, who runs the insurer's AI claims unit, called confirmed losses "almost certainly just the tip of the iceberg."

To make the losses visible, Allianz proposes sorting them into four groups. Direct losses come from the AI itself giving a wrong answer or making a bad decision. AI-related losses come from governance or disclosure failures, such as overstating what a product's AI can do. AI-accelerated losses are fraud, fake videos and cyberattacks that AI makes cheaper or larger. Dependency losses start with an unrelated problem but grow because a business cannot run without its AI tools.

Its dependency example is a large retailer that lost access to its AI forecasting and inventory systems during an outside cyber outage. Staff fell back on paper processes, and the wrong products shipped late.

Why one faulty model could hit many companies at once

The point that should worry investors most is correlation. Allianz notes that businesses across nearly every industry now lean on the same few large AI models, the same training data, and the same cloud providers and data centers. Michael Bruch, who heads its risk consulting advisory, warned that one error in a model or its data could cause losses "across multiple sectors."

Insurers fear this pattern more than any other. Their business works by spreading many unrelated risks across many customers. A shared failure behaves more like a hurricane, landing on thousands of policyholders on the same day. Allianz adds that because AI works on probabilities, mistakes will happen even at careful companies, and even a small error rate can quickly add up to many serious problems.

This is a quieter version of AI risk than the debate over runaway superintelligence. It is the near-term kind: a single flaw in a widely used system, showing up at once on many balance sheets.

Exclusions are spreading, and buyers may not notice

The US market is already reacting. In January, Verisk's ISO unit, which writes standard policy wording that many insurers adopt, released optional clauses that remove general liability coverage for injury or damage arising from generative AI. A review of state filings found 2,369 such exclusion records in force across 49 states by mid-2026, Insurance Business reported. W.R. Berkley has introduced an outright AI exclusion across its D&O, errors-and-omissions and fiduciary liability products, according to Insurance Journal.

A 2026 RAND Corporation study found carriers split three ways. A minority cover AI losses explicitly, a growing number exclude them, and most policies say nothing at all. Silence leaves the fight over who pays until after a loss.

Allianz says it is not taking a blanket approach. Bernard McNulty, its chief agent in Canada, told a risk managers' conference in Quebec City that the company has no "broad-brush exclusionary approach" and assesses AI exposure client by client.

Keep the incentives in view. Allianz sells this cover, and a system that makes AI losses visible also helps it price them. The four categories are its own proposal; no regulator or rival insurer has adopted them. Its $100 million figures are scenarios, and its release gives no count of AI claims or total paid.

For buyers, the practical step is narrower. Before the next renewal, check which policy would actually pay if an AI tool fails, and whether a new exclusion has slipped into the wording. Allianz says companies that invest in human oversight and a working fallback for when their AI goes down present a different risk, and that underwriting will increasingly reflect it.

Written by AI from the public sources linked below.