This week is about people, not products. An officer on a bridge with one colleague fewer. A machine driver in a container yard who has learned to ignore an alarm that goes off for nothing. A tanker crew that knows the camera is on. A ferry captain who used to do the tide sums in his head. In every story below, somebody bought or built a clever system. In every one, the real question is what it does to the person standing next to it. ABS put a number on that question this week, and it is the most useful number we have printed in a while, even though it comes from a computer model and not from a ship. Read the rest with it in mind.

The week's most important developments in shipping & oceans — distilled into a 5-minute read.

🔗 Quick Links

• 🧭 ABS: take one person off the bridge and the officer is overloaded in 61% of runs, not 8% — Digital Ship

  A computer model of one port approach, run 100 times for each crew size. Not measured on a ship, and ABS says so. Still the clearest warning of the year about cutting people before testing the work

• 🏗️ Busan terminal uses crane cameras to warn drivers when a worker is in their lane — Port Technology

  28 cameras on 14 cranes, about 70 yard machines. Workers carry no tag and are not identified. No accident numbers yet

• 🛰️ Lloyd's List Intelligence uses a satellite picture to check if a ship was where it said — Smart Maritime Network

  Software can only say a ship is probably faking its position. A picture of empty sea is stronger. It does not show where the ship went. No customer, price or coverage figure given

• 🇸🇬 Singapore will register a ship in minutes, not three days, and writes AI into its law — Smart Maritime Network

  The registry change is planned for 2027 and nobody says what the AI decides. The new law gives the port authority power over AI, autonomous vessels and drones

• 🎥 ShipIn says Prime Tanker Management saw 76% fewer safety gear breaches in 60 days of cameras — Smart Maritime Network

  Numbers from a named manager, published by the camera maker. The release does not say how they were counted, and people behave when they are watched

• 🇨🇳 China now has a national rulebook for ship data — Splash247

  In force since 1 October. Voluntary. Written by the state yards, COSCO and China Merchants, who will also be the ones using it

• ⚖️ AiatSea: the machine works. Can the person next to it still cope? — AI at Sea

  Four people in this week's news had their work changed by a system. Only one story tells us that anybody asked the person. Three questions to put to any supplier

🚀 Big Moves This Week

1. One Person Fewer on the Bridge, and the Officer Is Overloaded Seven Times as Often

ABS, the American class society, has published a paper with a simple message for owners: before you automate a ship or cut the crew, test on a computer whether the people who are left can still do the work.

The method is called human performance modelling. In plain words, you build a model of the tasks a person has to do, how long each one takes and how many can be handled at once, and you run a situation through it many times to see where the person runs out of capacity. Aircraft makers, nuclear plants and navies have used it for years. Shipping mostly has not.

The example is the reason to read it. ABS modelled one port approach and ran it 100 times for each crew size. With four people on the bridge, the officer on watch went over the model's overload limit in 8% of runs. With three people, in 61%. The officer's average peak workload score rose from 18.4 to 27.6. Same ship, same port, same equipment. The only change was one person fewer.

Now the honest part, and ABS says it first. These numbers are an illustration. They come from a model, not from a real bridge. A model is only as good as the task list and the timings put into it, and ABS states that modelling cannot replace simulator trials or the judgement of experienced people. Nobody should quote 61% as a fact about ships.

So why lead with it? Because it turns a feeling into a method. Every master knows the approach where one more thing would have been too much. Until now that was an opinion, and opinions lose against a spreadsheet that shows a crew saving. A workload model gives the safety side a number of its own to put on the table.

The paper covers more than manning. It names bridge design, alarm management, emergency procedures and remote operation centres, including the question of how many ships one person ashore can really watch. ABS asks for three things: put this modelling into the safety management system, build shared task lists and scenarios for ships, and agree a common way to check the results.

What to do with it. If a supplier or a yard offers you automation that "allows reduced manning", ask to see the workload study behind that sentence. If there is none, you have learned something. [Digital Ship · ABS · white paper · AiatSea, 27 September]

2. A Busan Terminal Teaches Its Cranes to Watch for People on Foot

At Busan New Container Terminal in South Korea, the software company CyberLogitec has installed a system that warns the drivers of yard machines when a worker on foot is close to them.

The problem is old and deadly. The cargo-handling safety body ICHCA ranks people struck by vehicles as the second biggest cause of death in cargo handling. In August, three industry bodies together asked terminals to do more about it.

How it works. There are 28 AI cameras on 14 quay cranes, looking down at the yard. They pick out people on foot and work out where each one is standing. About 70 yard machines carry precise satellite positioning. The system puts people and machines on the same live map and measures the distance between them. The driver gets a warning on a screen in the cab at 15, 10 and 5 metres, with a bar showing the direction.

Three choices here deserve credit. First, workers carry no tag. Most systems only protect people who remembered to wear the device. Second, the system finds a person but does not identify who it is, so it is a safety tool and not a tracking tool. Third, it only warns when the machine and the person are in the same lane. That last one matters most. As CyberLogitec's Dr Junhee Cho puts it: "The measure of a safety system is not how much it catches, but whether people keep trusting it." A driver who hears ten false alarms a shift stops listening by the third.

What is missing. There is no number for accidents or near misses, before or after. The accuracy figures, under 3 cm for machines and under 20 cm for people, come from "field validation", and the announcement does not say who did that validation. We do not know how long it has been running.

Why a ship manager should care. Your crew and your lashing gangs walk in that yard. And the design lesson travels well: any alarm on a bridge or in an engine room has to be believed to be worth anything. Ask your suppliers how many false alarms their system gives per watch. Few can answer. [Port Technology]

3. "Probably Lying" Becomes "Here Is the Picture"

Lloyd's List Intelligence has added a feature called Satellite Confirmed Spoofing to its Seasearcher ship tracking service.

First, the problem in plain words. Ships broadcast their own position. That broadcast can be faked, which is called spoofing. Two weeks ago we wrote about four tankers in the Black Sea that were broadcasting positions in the middle of Lima, Peru. Software can spot patterns like that and raise a flag. But as the company's research chief Arne Staal says, "detection models are by nature probabilistic". The software is giving you its best guess.

What is new. When the software flags a ship, the service can now look for a satellite picture of that spot at that time. If the picture shows no ship in the search area, the case moves from "probable" to "confirmed" spoofing.

One limit to be clear about. The picture shows the ship was not where it said. It does not show where the ship went. The company itself calls it an extra layer of confidence, not a final answer.

Why that difference is worth money. Think of the person who has to act on it. A chartering manager who refuses a ship, a bank that stops a payment, an insurer that questions a claim. "Our model says 80% likely" is hard to defend when the other side calls a lawyer. A picture of empty sea, with a time on it, is evidence that the position was false.

What we do not know, and it is a lot. No customer is named. No price. And the big one: how often is there a picture at all? Satellites do not photograph every piece of sea every hour, and clouds get in the way of ordinary cameras. A check that works for one flagged ship in three is very different from one that works for one in fifty. The announcement does not say.

The wider pattern. Last month a P&I club offered members a second way to check their own ship's position. This week a data company offers a second way to check somebody else's. Position is no longer something you simply believe. Before you sign for any screening service, ask what share of its spoofing flags it can confirm with a second source. [Smart Maritime Network · Lloyd's List Intelligence · AiatSea, 27 September]

4. Singapore Puts AI Into Its Ship Registry and Into Its Law in the Same Week

Two announcements came from Singapore's Maritime and Port Authority this week. Together they show a regulator using AI itself and taking the power to control it.

The registry. From the first half of 2027, registering a ship under the Singapore flag should take a few minutes for a standard application, down from up to three working days. A new system will cover the whole path from application to certificate, and owners and agents will be able to follow progress on screen. AI will "support the vetting and processing" of applications. The authority will invite companies to trial it before launch. The registry is not small: more than 4,400 ships and a record 137 million gross tons at the end of 2025.

The law. On 6 October a Bill went to Parliament to update the Act that set up the port authority in 1996. It would give the authority a legal base to regulate AI used in maritime operations, autonomous vessels and drones. It also creates a cyber security framework for the sector and gives the authority stronger powers to demand and check information. The second reading is in November.

Why put these two together. Most regulators are still writing discussion papers about AI. Singapore is doing two concrete things at once. It is trusting AI with part of its own paperwork, and it is making sure it can set rules for everyone else's.

Our caveats. The registry system is a plan, not a working service. "Support the vetting" could mean the AI checks that a form is complete, or it could mean the AI decides who gets a flag. Those are very different, and nobody has said which. If an application is refused, will a person explain why? On the Bill, the published summary gives powers but no detail and no penalties. The real rules will come later, in the regulations made under it.

What to do. If you run ships through Singapore, read the Bill before November and ask your association what it plans to say. And if you register ships there, ask to join the trial. It is the cheapest way to find out what the AI does with your application. [Smart Maritime Network · Splash247 · Digital Ship]

5. A Camera Maker Publishes a Customer's Numbers. Now Ask What They Count.

Prime Tanker Management has put ShipIn's FleetVision cameras across its managed fleet. ShipIn, the company that makes the cameras, has published what changed over a 60-day comparison period.

The numbers. Breaches of safety gear rules down 76%. Bridge-related breaches down 51%. Technical breaches down 69%. Crew taking part in drills, toolbox talks and briefings up 3.4 times. The company's score on ShipIn's own scale went from 82 to 94. Masters use the footage with crews in safety meetings, and the office uses it to compare ships and to show oil majors how procedures are followed on board.

First, credit where it is due. These are figures from a named manager with a stated period, even if the camera maker is the one putting them out. For weeks we have asked camera companies and their customers to do that. Last week one manager published a number from a single ship. This is a fleet. The direction is right.

Now the question we asked on 30 August, when a serving master mariner raised it in print. Who counted, and how? The release comes from the company that sells the cameras. It does not explain how a breach was counted or what the earlier period was. As far as we can tell, the count comes from the camera system itself. And when people know a camera is on, they put the helmet on. That is a real change in behaviour, and it may well save an injury. But it is not the same as a safer ship. A crew can learn what the camera looks for much faster than a company can change its safety culture.

What would settle it. Numbers the camera does not produce: injuries, near misses reported by the crew themselves, port State control findings, oil major inspection results, insurance claims. If those move the same way over a year, the case is made. Neither company has published them, and the fleet size is not given either.

A fair word on repetition. This is the fifth time since August that ShipIn appears in this newsletter. We run it because it brings new numbers from a customer, not because it is a new launch. We will not run a sixth without an outcome that somebody other than the camera has counted.

For managers who already have cameras: ask your own safety department one thing. Since the cameras went in, have crew-reported near misses gone up or down? If they went down, find out whether the ship got safer or the crew got quieter. [Smart Maritime Network · ShipIn release · AiatSea, 30 August · AiatSea, 4 October]

6. China Writes the Dictionary for Ship Data

On 1 October a new Chinese national standard came into force. Its number is GB/T 47313-2026 and its subject is how to classify, grade and label the data on an intelligent ship.

What that means in plain words. Every modern ship produces data from hundreds of pieces of equipment, and each maker names and stores it in its own way. That is why joining two systems on board is slow and expensive. This standard sets one common way to sort ship data into types, to give each item a code that other systems can recognise, and to grade it by how much harm it would do if it leaked or was changed. The grade then decides how strongly it must be protected. It covers the whole life of the ship, from design and building to operation, including equipment connections and onboard software.

Who wrote it. Research institutes of China State Shipbuilding Corporation, COSCO Shipping, China Merchants Industry, a transport ministry institute, two maritime universities, and other yards and port groups. In other words, the people who build most of the world's ships, and some of the people who own them.

Why this answers a question we asked. Three weeks ago China Merchants launched an operating system for ships, and we asked who owns the digital layer inside a Chinese-built ship. A standard like this is the other half of that plan. First you agree the language. Then you sell the system that speaks it.

The limits. It is voluntary. There is no list of yards or equipment makers that have promised to follow it, and no ship named as built to it. A standard on paper changes nothing until it shows up in a building contract.

And the question the standard does not answer. As described, it covers how data is named and how sensitive it is. Naming data does not settle who owns it. For an owner ordering in China, that is still the point to settle. Put three lines in the building contract: who owns the data, where it is stored, and your right to export all of it in a format another supplier can read. If the yard says it follows the new standard, good. Ask which grade your engine and position data fall into, and who decided. [Splash247 · AiatSea, 20 September]

7. AiatSea: The Machine Works. Can the Person Next to It Still Cope?

Four people appeared in this week's news without being named. It is worth looking at each of them.

The officer on watch. In the ABS model, nothing on the bridge got worse. One colleague left. The officer went from overloaded in 8 approaches out of 100 to overloaded in 61. No screen failed and no sensor lied. The work of the fourth person did not disappear when that person did. It landed on someone.

The driver in the yard. At Busan, the engineers spent their effort on something that sounds small: making the alarm stay quiet unless a worker is in the same lane. They did it because they know what a driver does with an alarm that cries wolf. He turns it down, or he stops hearing it. The clever part of that system is not the camera. It is the respect for the driver's attention.

The crew under the camera. On Prime Tanker's ships, safety gear breaches fell by three quarters in 60 days. Good. But think about the seafarer for a moment. He now works with something watching that scores him. Maybe he feels safer. Maybe he feels judged. Maybe he reports fewer of his own small mistakes, because the office already sees enough. The numbers published this week cannot tell us which.

The captain at St Helier. The fourth person is not from an AI story at all, and he is the most useful one. This summer the Stena ferry Stena Vinga tested a new kind of electronic chart on the run to Jersey. It shows depth, tide height and current together, live, on one screen. St Helier has a tidal range of up to 11 metres, and Captain Johan Karlsson says of the harbour, "the vessel almost doesn't fit inside". His verdict on the new display was plain: the sums he used to do in his head are now in front of him. The supplier added something honest. All that information existed before. It just took a lot of manual work to use.

Put the four side by side. In one case a system took a person away and left the work. In one it added an alarm and thought hard about whether the person would trust it. In one it added a watcher. In one it took work out of a captain's head and put it on a screen, and they asked him what he thought. Only in the last story are we told that anybody asked the person. The others may have been asked too. If so, nobody thought it worth saying.

That is the difference we want readers to look for. Vendors sell what the machine can do. Almost none of them tell you what it does to the person next to it: more to watch, less to do, another alarm, another score. That is not in the brochure, and until this week there was no accepted way to measure it. ABS has now pointed at one.

Three questions for your next AI purchase. Whose work changes on the day this goes live, by name and by rank? Does it add tasks, alarms or screens to that person, and what does it take away in return? And did anybody ask that person before the contract was signed? If the answer to the last one is no, you are not buying a tool for your crew. You are running a test on them. [Digital Ship · Stena Vinga · Digital Ship · ABS · Port Technology · Smart Maritime Network]

📊 Why It Matters — Strategic Impact Table

🔭 On Our Radar

• 🗳️ Who Is Actually Paying for Shipping's AI Accounts? — carried for a ninth week: adoption figures still count companies and not people, our reader survey is still open, we monitor whether any manager, union or class society publishes guidance on who provides and pays for AI tools at work, and we will publish our own findings when we have enough answers, including if they prove us wrong.

• 🧭 Does Anyone Run the Workload Model on a Real Bridge? — ABS showed overload rising from 8% to 61% in a computer model, we monitor whether any owner, yard or simulator centre repeats the test with real officers, and track whether class or flag States begin asking for a workload study before they accept a lower manning level.

• 🏗️ Will Busan Publish an Accident Number? — the terminal has cameras and warnings but no before and after figures, we monitor whether the terminal, its insurer or a port safety body publishes near miss or injury data, and track whether other terminals copy the no-tag, same-lane design.

• 🛰️ How Often Is There Actually a Picture? — a satellite image can confirm a faked position only when a satellite was looking, we monitor whether Lloyd's List Intelligence or any rival publishes the share of flags it can confirm, and track whether banks and insurers start asking for image proof in sanctions cases.

• 🇸🇬 What Does the AI Decide in the Singapore Registry? — the plan says AI will support the checking of applications, we monitor what the trial shows about the AI's real role, whether a person signs off every refusal, and what the November reading adds on AI and autonomous vessels.

• 🎥 Does Any Camera Customer Publish a Number the Camera Did Not Count? — two managers in two weeks have had results published with no outside count, we monitor whether Prime Tanker, Acheon Akti or any insurer publishes injuries, crew-reported near misses or claims over at least six months, and we hold further camera coverage until one does.

• 🇨🇳 Does the Chinese Data Standard Reach a Building Contract? — the standard is voluntary and no yard has yet said it applies it, we monitor which yards and equipment makers adopt it, whether it appears in newbuilding specifications, and whether owners keep the right to take their data out.

📅 Critical Maritime AI Research Areas for Managers

1. 🧭 Workload Modelling Checked Against Real Bridges: the ABS figures come from a model with no ship behind them. Research should run the same port approaches in a full simulator with serving officers at different crew sizes, and publish how close the model comes to what people actually do.
2. 🏗️ False Alarms and Trust in Safety Systems: a warning that sounds for nothing teaches people to ignore it. Research should measure false alarm rates for AI safety systems on bridges, in engine rooms and in terminals, and find the rate at which crews stop responding.
3. 🎥 What Cameras Do to Crew Reporting: breach counts fall when cameras go in, but nobody has published what happens to voluntary near miss reports. Research should compare crew reporting before and after camera installation across several fleets, using data the camera vendor does not hold.
4. 🛰️ Confirmation Rates for Spoofing Detection: detection is a probability and confirmation needs a second source. Research should measure how often satellite images, radar or other sources can confirm a flagged position, by region, weather and time of day.
5. 🇸🇬 AI Inside Flag and Port State Administrations: a major registry will use AI to process applications. Research should set out which registry decisions can safely be automated, which need a person, and how an owner can challenge a decision made with machine help.
6. 🇨🇳 Data Grading Standards and Owner Rights: a national standard now grades ship data by sensitivity. Research should compare it with the international ship data standards, and test whether grading helps or hinders an owner's right to move data between suppliers.

📈 Top Investment Opportunities

1. 🧭 Workload Testing as a Service — a class society has said that automation and manning changes should be tested for crew workload first, and almost nobody in shipping sells that test — the opportunity is in modelling services for owners and yards, shared task libraries for ship types, and simulator centres that can check a model against real officers — ABS · Digital Ship
2. 🏗️ Tag-Free Protection for People Around Heavy Machines — a working terminal now locates people on foot without a device and without identifying them — the opportunity is in the same approach for smaller terminals, shipyards, ro-ro decks and offshore decks, and in the measurement of false alarms that decides whether drivers keep trusting it — CyberLogitec · Busan
3. 🛰️ Second-Source Proof for Position Claims — screening is moving from a probability to evidence a lawyer can use — the opportunity is in satellite imaging with frequent revisits, radar imaging that works through cloud, and services that package the proof for banks, insurers and charterers — Lloyd's List Intelligence
4. 🇸🇬 Software for Regulators — a leading flag is cutting registration from days to minutes and taking legal power over AI at the same moment — the opportunity is in registry and licensing systems for other flag and port States, and in tools that help owners show compliance with the technology rules that follow — Singapore MPA
5. 🎥 Independent Measurement of Safety Technology — camera results still come from the camera maker — the opportunity is in insurer-backed or third-party studies that link safety systems to injuries, inspections and claims, and in crew reporting tools that sit outside the vendor's system — Prime Tanker + ShipIn

📅 Top Monthly Picks

1. 🛰️ Four Tankers "in Lima" — Black Sea tankers broadcast positions from the middle of Peru, and the startup that caught it argued that normal data cleaning throws the evidence away; this week a satellite picture became the way to prove such a case, which is the next step in the same story — Splash247
2. 🎓 Should Every Officer Be Trained to Doubt the Machine? — a training company chief asked the IMO to make AI literacy part of officer certification; this week ABS added the other half, which is whether the officer has any capacity left to doubt anything — Seatrade Maritime
3. 🎥 A Ship Manager Publishes a Number for Camera AI, From One Ship — bridge near misses down almost 20 times on a single pilot vessel, with no word on who counted; this week a second manager published fleet numbers, and the same question is still open — Splash247
4. 📋 Almost Everyone Uses AI, Almost Nobody Has Rules for It — a survey found 63% of maritime staff use AI every day and only 8% say their company governs it, from a sample of 60; this week Singapore started writing the rules into law instead of waiting for companies to do it — Smart Maritime Network
5. 🔌 China Merchants Wants to Be the Operating System Under Your Ship — a state-backed group launched a platform from chips to applications with no customer named; this week the national data standard behind that kind of platform came into force, written partly by the same group — Smart Maritime Network

Have feedback or a story tip? Email us or connect on LinkedIn.