Total losses have fallen by nearly two thirds in a decade. In the same window one container ship cost the market $2.8bn, the largest marine loss ever recorded. Frequency and severity have come apart, and the two halves of the market disagree about which one to price.
At 1.24am on 26 March 2024, the container ship Dali lost electrical power in the Patapsco River, about half a mile from the Francis Scott Key Bridge. The crew had roughly four minutes.
Six road workers on the bridge did not get off it.
Every underwriter on that placement knew the vessel, the route and the limits. What none of them had priced was a bridge.
Two years later the arithmetic closed, and it closed a long way from where it started.
Early estimates put the insured loss somewhere near $1.5bn. In an analysis published on 30 April 2026, Howden Re put the settled figure above $2.8bn.
Hugo Chelton, a Managing Director at Howden Re, made the point that gets lost in the headline number. The gap between the first estimate and the final bill was itself larger than most marine catastrophes: “A $1.3 billion deterioration would be a major loss event in its own right.”
That makes Baltimore the largest single marine insured loss ever recorded. The previous benchmark was the Costa Concordia at roughly $1.6bn, and that was 2012.
The money moved in pieces rather than one payment, which is why the figures in circulation disagree with each other. The State of Maryland settled with the vessel’s owner and operator, Grace Ocean and Synergy Marine, for $2.25bn - a number worth taking from the Maryland Attorney General’s own announcement rather than from the several outlets that rounded it to $2.5bn. Separately, Chubb settled as the bridge’s insurer. Pollution liability, wreck removal and lost toll revenue made up the rest.
Almost none of it stayed with the primary market. A loss of that size runs up through the International Group of P&I Clubs’ pooling arrangement and into its reinsurance tower, which carries a limit of around $3bn. The exposure concentrated where it usually does, with the large reinsurers and the retrocession market behind them.
One ship, one loss of power, four minutes, and the reinsurance tower for the entire mutual P&I system was in play.
The National Transportation Safety Board’s final report, published in November 2025, traced the blackout to the vessel’s electrical system and pointed at how it was built, and claims against the shipbuilder are still live. For an underwriter that detail matters more than the total: the proximate cause was not seamanship, weather or a route decision. It was a design fault that had been aboard since delivery, invisible to every renewal questionnaire the ship ever answered.
Set against that, the safety record is genuinely good, and it has been improving for a decade.
Allianz Commercial published its Safety and Shipping Review on 24 June 2026. It counts 43 total losses in 2025. Across 2021 to 2025 the industry lost 350 vessels, an average of 70 a year, against 555 across the previous five years, an average of 111.
Total losses have fallen by nearly two thirds in a decade
Vessels over 100GT declared a total loss, per year
Allianz Commercial, Safety and Shipping Review 2026
Incidents short of a total loss fell too, from 3,353 in 2024 to 2,818 in 2025, a 16% drop in a single year.
“Our analysis shows the shipping industry has made significant improvements in maritime safety in recent years,” said Thomas Lillelund, chief executive of Allianz Commercial.
He is right, and it is the least useful sentence in the report for anyone pricing the line. Baltimore happened in a year when the fleet was safer than it had ever been. Frequency and severity have come apart, and only one of them is falling.
Ask which peril to worry about and the trade press will answer war risk, or battery fires, or sanctions. The Allianz incident data says something duller.
Machinery damage or failure caused 1,505 of the 2,818 incidents in 2025. More than half of everything that went wrong at sea last year was an engine, a generator or a pump.
Collisions accounted for 260.
Nothing about that is dramatic, which is roughly the point. The most expensive marine loss in history began as a machinery failure, and machinery failure is also the most common thing that happens to a ship. The tail and the body of the distribution have the same cause.
Geography concentrates it. The East Mediterranean and Black Sea recorded 622 incidents, the British Isles 619 - two regions accounting for nearly half the global count between them, one because of a war and one because of traffic density and weather.
A newer failure mode does not appear in the decade of incident data at all, because until recently it did not exist. Researchers from GPSPatron and Gdynia Maritime University measured satellite navigation off Gdansk between June and October 2025 and found GNSS positioning unavailable about 17% of the time at the summer peak, with every recorded event combining jamming with spoofing: a receiver that is not just blinded but lied to.
At least two collisions and groundings in 2025, across the Baltic, Hormuz and the Red Sea, have been linked to that interference. In January 2026 thirteen coastal European states and Iceland issued a joint statement on it.
It rhymes with the Dali. The blackout was a fault no renewal questionnaire could see, because it was built into the ship. Jamming is a fault no questionnaire will see, because it is built into the route.
Fire is the peril where the trend is going the wrong way.
There were more than 200 fires aboard large vessels in 2025. That was down on 2024 and still the second highest total of the decade, and at least nine of those fires destroyed the ship. Allianz puts a container ship fire somewhere in the world every 17 days.
On 3 June 2025 the car carrier Morning Midas was three weeks out of Yantai, bound for Mexico with more than 3,000 vehicles aboard. Around 700 of them were electric or hybrid. Smoke came off the EV deck, the crew ran the onboard suppression system, the fire kept going, and all 22 of them left in a lifeboat. The ship was insured in part by Steamship Mutual.
A lithium-ion fire in a sealed vehicle deck is not a fire the ship was designed to fight. Conventional systems smother a fire by cutting off oxygen, and a battery in thermal runaway supplies its own.
That shows up in the mutual sector’s accounts rather than in a rating table. P&I claims reached a ten-year peak, and after two years close to breakeven the clubs’ 2024-25 financial year produced a combined ratio just under 110%.
The response was a general increase averaging around 6% across the International Group for 2026, with seven of the twelve clubs asking 5% and the American Club and Steamship both at 8%.
Two clubs did the opposite in the same year. Britannia returned 5% of members’ premium and Gard returned 10%. Free reserves across the group rose 4.81% to $5.96bn, though measured per tonne of entered tonnage they remain below where they sat before 2020.
The Allianz review contains one figure that reads less like a statistic than a hostage note. At the point of writing, roughly 1,150 cargo-carrying vessels over 100GT were waiting to leave the Persian Gulf: 29 million gross tonnes, about 20,000 seafarers, and a combined vessel and cargo value near $125bn.
War risk cover is written per voyage, so a corridor can reprice in days.
Broker figures reported by Al Jazeera on 23 July 2026 put war risk on a Hormuz transit at 7.5% to 10% of hull value, against 1% to 3% before the escalation. On a $100m tanker that is the difference between a few hundred thousand dollars and eight figures, for one passage.
Hormuz repriced by an order of magnitude
War risk premium as a share of hull value
Broker figures reported by Al Jazeera, 23 July 2026. War risk pricing moves weekly and is negotiated per voyage, so treat these as a snapshot.
Traffic answered the price. Hormuz was carrying 120 to 140 vessels a day before the escalation and was down to between 10 and 16 in late July. Freight for a Gulf to China crude cargo hit $77.96 a tonne against a five-year average of $18.91.
Marcus Baker, Marsh’s Global Head of Marine, Cargo and Logistics, described the market’s posture drily: insurers “are charging a bit more for risk in the Red Sea”. Simone Krummaker, Associate Professor of Insurance at Bayes Business School, made the more important observation, which is that cover has stayed available on most of these voyages. The market is repricing, not withdrawing.
Captain Rahul Khanna, who runs marine risk consulting at Allianz Commercial, framed the underwriting problem as one of routes rather than perils: “We are seeing growing uncertainty around shipping routes. Any type of event - a conflict, pandemic or a grounded vessel blocking a key port or shipping canal - can potentially cause a major disruption to shipping and supply chains.”
Every number in this piece so far describes the insured fleet. There is another one.
Allianz reckons around 17% of the world’s tankers now belong to the shadow fleet: vessels moving sanctioned oil outside the mainstream system, close to 600 of them trading Russian crude alone. They run under flags of convenience, switch registration when scrutiny arrives, and in many cases sail without liability cover. They are also old. The average age of the global fleet rose to 23 years in 2025, and the shadow fleet sits at the aged end of it.
Nobody agrees how big the shadow fleet is
Estimated shadow fleet tanker counts, by source
Allianz Commercial Safety and Shipping Review 2026; S&P Global Commodities at Sea; analyst estimates and the Ukrainian intelligence catalogue as reported in early 2026. The counts measure different definitions, which is the point.
The counts disagree because the definitions do. S&P Global counts by behaviour and reaches 978 tankers, roughly 18.5% of the global oil tanker fleet. Ukraine’s intelligence services keep a catalogue of named ships and had 1,337 on it by February 2026. Analyst estimates run 600 to 800. When four serious counters disagree by a factor of two, underwriters are not pricing a risk. They are pricing a rumour.
Allianz counts tens of incidents already involving shadow vessels: fires, collisions, spills. The insurance problem is not the shadow tanker itself, which by definition carries no policy that will respond. It is the insured ship on the other side of the collision, whose hull insurer pays in full and then discovers there is nobody to recover from.
The size of the market is smaller than the drama suggests. IUMI’s most recent Stats Report, published on 3 November 2025, puts the global marine premium base at $39.92bn, up 1.5%.
Cargo is more than half of the marine premium base
Global marine premium by segment, 2024, with year-on-year change
IUMI Stats Report 2025, covering 2024. The 2026 edition is not due until around November.
Worth being clear about the vintage: that is 2024 data, and IUMI does not publish the 2025 figures until late in the year. Anyone quoting a 2026 marine premium number is quoting a forecast.
Cargo carried $22.64bn of it and its loss ratios improved for a sixth consecutive year. Hull grew 3.5% to $9.67bn. Offshore energy went the other way, down 7.9% to $4.34bn, with London still writing about 60% of it. “The ocean hull and cargo markets have been relatively stable,” IUMI Secretary General Lars Lange said of the year.
Cargo’s improving loss ratios sit oddly beside its crime ledger. The BSI and TT Club cargo theft report for 2025, published in April 2026, puts Brazil at 22% of recorded thefts, Mexico at 15%, the United States at 13% and India at 11%, with food and drink the most stolen commodity and trucks involved in 70% of incidents. Most of what a marine cargo policy pays for now happens on a road. Insiders featured in 22% of thefts globally, US rail theft went from 4% to 10% of incidents in a single year, and the growth areas are fictitious pickups and double-brokering: fraud committed with paperwork, not bolt cutters.
Stable is where the disagreement starts. Howden Re observed that despite Baltimore, parts of the market softened by 15% to 20% at the April 2026 renewals, with new capital keeping pressure on price. The mutual clubs, looking at the same calendar and a 110% combined ratio, went up 6%. One half of the market is pricing a decade of falling frequency; the other is paying for severity that has already arrived.
Both can be right, because they are not insuring the same thing. That is also what makes the line hard to run on software designed for a stable book: a rating engine whose tables can be changed by an underwriter rather than a release cycle, war risk written per voyage instead of per year, a claims file that can hold a two-year deterioration without being reopened as a new matter, and an audit trail good enough to show a reinsurer why the number moved. Marine cargo and specialty lines generally punish platforms that assume last year’s product is this year’s product.
Justus Heinrich, who leads marine hull product at Allianz Commercial, put the underwriting problem in one line: “Insurance markets react quickly to crises, but the real challenge for companies is understanding how risks are interconnected.”
Baltimore was one ship, one electrical fault and four minutes, and it cost the global marine market more than offshore energy collects in premium in half a year.
There is no institutional premium series for cannabis insurance, and the forecasts that fill the gap disagree with each other. What is measurable says something the growth reports miss: the number of licensed businesses has fallen 13% in two years while property rates rose up to 40%.
Only 11% of agentic AI use cases reached production last year, and the industry treats that as the problem. Cigna's claim-denial system reached full production and averaged 1.2 seconds of human review. What Zurich, IAG, WTW and Lemonade did differently, and what a court in Minnesota decided it wanted to see.
Insurers spent 2025 and 2026 writing AI out of standard policies and building a new market to replace it. Here is what Armilla, Munich Re, AIUC, Klaimee, HSB and Testudo actually sell, and what it takes to launch a product like theirs.
Book a live, personalized demo with our product team - tell us your use case and see the platform work with your data. No commitment.