Last week we asked who checks the checkers. This week somebody answered, and the answer was: whoever the company pays. On Friday Anthropic named its first outside AI inspector, with desks and access inside the company, and said openly that it will pay the bill itself because nobody else will. Shipping knows this arrangement well. The owner pays the class society, and we have spent 150 years building flag States, port State control and insurers around that fact so it does not matter too much. Read this week's news with that in mind. One number this week was checked by a government body with nothing to sell. One approval was called a landmark by the company that paid for it, while the certificate itself says non-essential. One port refused to sign until the supplier proved it on the port's own operation. The difference between those three is the whole story of buying AI in 2026.

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

🔗 Quick Links

• 🚢 Shipping Technology gives its inland pilot the engine as well as the rudder — Smart Maritime Network

  Captain sets destination and arrival time, the system handles course and power. Runs on 200 vessels. The 5% saving was checked by Rijkswaterstaat, the new 10% is the vendor's own

• 🛰️ Galileo produces the first civil position protected against spoofing — ESA

  Five satellites, two hours, a live spoofing test in Norway, and the protected receiver held the true position. Service due 2027, free worldwide

• 🏗️ Yilport reports 21% more crane moves per hour with Kaleris software — Splash247

  Measured over three months on 31 cranes at Gebze. Dispatchers keep the final say. Now going to three more terminals

• 🇮🇳 JNPA signs a $9.7m AI digital twin only after a live test — Splash247

  India's largest container port made bidders prove it on real operations first. That should be normal. It is not

• ⚙️ TS Cyanergy calls it a landmark, ABS calls it non-essential — Smart Maritime Network

  A wireless engine-monitoring approval with a real review behind it. Read what it covers before you read the headline

• 🔌 China Merchants launches a vessel operating system — Smart Maritime Network

  One platform from chips to apps, so equipment from different makers can share data. No customer or vessel count given

• ⚖️ AiatSea: who pays the inspector? — AI at Sea

  An AI lab hired its first outside inspector and pays it directly. Shipping has lived with that exact problem for 150 years, and has answers the AI industry does not

🚀 Big Moves This Week

1. The Inland Autopilot Gets the Engine, and One of Its Numbers Was Checked by Someone Else

Rotterdam-based Shipping Technology has launched ST Sailing Pro, an extension of its ST Sailing track pilot for inland vessels. Until now the system steered. Now it also controls propulsion.

It works like this. The captain casts off and manoeuvres out by hand. Then he or she switches the system on and enters a destination and an arrival time. From there the system keeps calculating the course and engine power needed to arrive on time, taking account of locks, currents, water levels and squat (the way a hull sinks lower in shallow water at speed). The company says the possible arrival window can move by up to three hours depending on conditions. The captain approves what the system proposes, such as whether to overtake, hold back, or arrive earlier or later.

This is not a lab product. Shipping Technology started collecting vessel data in 2018 and added automatic rudder control five years ago. Its equipment now runs on more than 400 vessels, and ST Sailing alone runs on about 200, with more than 315,000 sailing hours and 4.4 million kilometres logged. Propulsion control was tested first on a thruster-driven river cruise ship, then for six months on a single conventional vessel under a Smart Shipping permit, and is now in trials across dry cargo, tanker and passenger vessels. It works at level 2 of the Rhine Commission's automation scale: the captain supervises at all times and can take over at any moment.

Now the part that made us lead with it. The existing ST Sailing product claims a 5% fuel saving, and that figure was validated by Rijkswaterstaat, part of the Dutch Ministry of Infrastructure and Water Management, in certified field testing. That is a government body with nothing to sell, measuring a result on real vessels. After weeks of this newsletter asking for independently checked numbers, this is the cleanest one we have found.

And the part we must be honest about. The new product claims a further 10% on top of that 5%. That figure is the company's own, from trials, and it has not been through the same check. It may well hold. But the right way to read this announcement is: one number checked, one number promised. A buyer should treat them differently.

Why a deep-sea manager should care about river barges. Inland shipping is where "the machine drives, the human decides" is being tested at volume on ordinary commercial vessels with ordinary engines and gearboxes, not on showcase newbuilds. The idea of giving the system an arrival time instead of a speed is exactly the just-in-time arrival logic that ports and charterers keep asking ocean-going ships for. It is running in daily service on the Rhine first. [Smart Maritime Network · SWZ Maritime · MarineLink]

2. Galileo Holds the True Position While the Spoofer Lies

On the afternoon of 16 September, for two hours, five Galileo satellites over Europe transmitted an encrypted signal for the first time. Receivers at ESA's laboratory in the Netherlands and at Andøya in Norway used it to calculate a position. Europe's space agencies call it the first civil satellite position authenticated through both the navigation data and the ranging signal.

In plain words, what that means. A satellite position depends on two things: the message the satellite sends (where it is, what time it is) and the signal the receiver measures the distance with. Since 2025, Galileo's OSNMA service has let receivers check that the message is genuine. But the signal itself could still be faked, and that is how spoofing works: a transmitter copies the predictable signal and pushes a false position onto the receiver. The new Signal Authentication Service (SAS) protects the signal as well.

The test that matters. Much of the demonstration ran during Jammertest at Andøya, where controlled jamming and spoofing is done with real signals over the air. In the published results, an ordinary commercial receiver was pulled to false positions by the spoofer. The SAS-enabled receiver kept the true one.

What it is not, yet. It was a two-hour test with five satellites. Validation and accreditation run through 2027 before the service is declared operational. It will be free and worldwide. It does not stop jamming, which simply drowns the signal. It lets a receiver know when it is being lied to, which is the more dangerous problem.

Why this is in an AI newsletter. Almost every AI tool on a ship assumes the position is true: voyage optimisation, emissions reporting, collision-avoidance support, route compliance, AIS-based risk screening. Every one of them inherits a spoofed position without noticing. The practical question for your next newbuild or bridge refit is simple, and worth asking now: will the GNSS receivers support Galileo OSNMA and the E6 signal, and will the bridge show the crew when a position is not authenticated? A warning buried in a log file is no use at 0300 in the Gulf. [ESA · EUSPA · Splash247]

3. Yilport Gets 21% More From Cranes It Already Owns

Yilport says rubber-tyred gantry crane productivity at its Gebze terminal in Turkey has risen by 21% after it deployed Kaleris' RTG Optimization software.

The detail that makes this worth running is how it was measured: moves per hour, over the first three months of operation, across the terminal's 31-crane fleet. That is a named customer, a named metric, a stated period and a stated fleet size, which is more than most optimisation announcements give you.

The software looks at vessel, truck and yard demand all the time and recommends where each crane should work, to balance the load, cut idle time and reduce empty travel across the yard. The dispatchers keep control of crane assignments. The system recommends, a person decides. It sits on top of the Kaleris N4 terminal operating system Yilport already uses, and it can handle mixed fleets, including remotely operated cranes.

The best signal is what Yilport does next. It plans to take the software to Liscont and Leixões in Portugal and Gävle in Sweden, where the installations will include remotely controlled automated cranes. A customer rolling a product out to three more sites is a stronger signal than any single percentage.

Our caveats. The 21% is reported by the two companies, not by an outside auditor, and the baseline is not published. We do not know whether volumes or the vessel mix changed over the same three months. We ran Kaleris two weeks ago for an acquisition. We run it again because this time it is a customer result, not a deal.

Why it matters to ship managers. Crane productivity at the terminal is your time alongside. Every move per hour gained at the quay is berth time you do not pay for. It is also a clean example of the pattern we keep seeing where AI works: more out of the equipment you already have, with a human still making the call. [Splash247 · AiatSea, 6 September]

4. India's Biggest Container Port Made the Supplier Prove It First

India's Jawaharlal Nehru Port Authority (JNPA) has awarded SoftTech Engineers a Rs929.5m ($9.7m) multi-year contract to build an AI and machine-learning digital twin and a 24-hour command and control centre for the whole port.

The scope is large. The system will cover roughly 34 square kilometres, bringing sea side, land side and road traffic into one live picture. It will connect data across the port's five terminal operators, customs and rail, and cover vessel turnaround and just-in-time arrivals, container flows, gate queues and vehicle tracking. SoftTech leads design, AI development, integration, construction of the command centre and long-term operation, using its CIVIT platform, with Dassault Systèmes' 3DEXPERIENCE providing the core virtual-twin environment.

The sentence we care about most is near the bottom: the contract was won through a competitive process that included a live proof of concept. In other words, bidders had to show their system working on the port's own reality before anyone signed.

That is how every AI purchase should work, and very few do. Most maritime AI is still bought from a demonstration on the vendor's data, with the vendor's chosen example, measured by the vendor's chosen number. A live test on your own operation, with a success measure you set in advance, is the single most effective protection a buyer has.

Continuity note. In August the port appointed an adviser to shape this project. A month later it has a contract with a live test behind it. That is the path from advice to purchase done properly.

What it is not: a working system. It is a contract. The command centre, the five-operator data links and the customs and rail integration are all still to be built, and joining five terminal operators' data is where projects like this usually slow down. [Splash247 · Smart Maritime Network, August]

5. A Landmark Validation, Says the Press Release. Non-Essential, Says the Certificate.

TS Cyanergy, based in Houston, has received Product Design Assessment (PDA) approval from ABS for CE2, a wireless engine-monitoring platform for ships and offshore installations. It uses wireless sensors to collect engine performance, fuel consumption, emissions and operating data, plus maintenance warnings, and sends them to cloud analytics and real-time reports.

The review was real work. ABS looked at the technical design, tested the product, reviewed the documentation and evaluated its cyber security. That is worth something, and wireless sensing that avoids pulling new cable through an engine room is a sensible idea for older ships.

Now compare the two descriptions. The company's press release calls it "a landmark validation" of its wireless approach. The approval itself confirms the platform can be installed as a non-essential engine monitoring system aboard ABS-classed vessels.

Both are true. But they are not the same claim. "Non-essential" means the ship does not rely on it. It is not part of the alarm and monitoring system the class rules require, and it is not what the engineer depends on to keep the engine safe. It is an extra layer of information beside the systems that are.

We are running this because we promised to. Last week we asked whether any certifier ever publishes what it did not certify as clearly as what it did. Here the scope is in the approval, in plain words, and the marketing walked straight past it. That is a pattern every buyer should expect.

The practical rule. When a supplier shows you a class approval, ask for the certificate, not the press release, and read the scope line first. Also ask the two questions the approval does not answer: how accurate is the fuel and emissions data, and has it been checked against a calibrated flowmeter on a real engine? No customer, vessel count or accuracy figure has been published. [Smart Maritime Network · TS Cyanergy · AiatSea, 13 September]

6. China Merchants Wants to Be the Operating System Under Your Ship

China Merchants Group has launched an intelligent vessel operating system, developed by CM Marine Intelligence, a subsidiary of CM Energy. Its purpose is to solve a problem every superintendent knows: equipment from different makers that speaks different protocols, keeps its data in separate boxes, and does not talk to anything else on board.

The system combines an operating system, onboard networks, digital infrastructure and cloud-based training platforms. The aim is a common digital foundation so that equipment from different manufacturers can exchange data, and so applications for navigation, energy efficiency and vessel operations can run on top. The group's chairman describes it as a full-stack architecture spanning chips, operating systems and applications.

Why this is interesting. Shipping has talked about a common data layer on board for twenty years, and the industry groups have mostly produced standards documents. A state-backed group that owns shipyards, ports and a fleet can do something a standards committee cannot: it can install the thing on the ships it builds. The timing is worth noticing too. This same week, 20 new car carriers for CIDO Shipping under construction at China Merchants Heavy Industry were confirmed to carry a Korean autonomous navigation system. The yard builds the hull. Who owns the digital layer inside it is still an open contest.

What is missing, and it is a lot. No customer, no vessel count, no list of equipment makers who have signed up, no statement on which open standards it follows, and nothing on whether an owner can take the data out or run another supplier's application on it. "Chips to applications" is also a reminder that whoever owns the bottom of the stack decides what runs on top.

The question for owners ordering Chinese-built tonnage: if this system arrives with the ship, who owns the data it collects, where is it stored, and can you replace any layer of it without replacing all of it? Those answers are much easier to get in the building contract than after delivery. [Smart Maritime Network · Smart Maritime Network · CIDO · Splash247]

7. AiatSea: Who Pays the Inspector? The AI Industry Just Met Shipping's Oldest Question

Last week Dario Amodei, chief executive of Anthropic, asked the AI industry to slow down and promised one concrete step: outside evaluators working inside his company with access close to an employee's. On Friday he delivered it. Anthropic named Faculty, the AI business Accenture bought in January, as its first embedded evaluator. Faculty's people will test and try to break the models, check how they behave, test the safeguards, and report incidents. Both companies say they expect to invest at least $1 billion each over five years in this kind of work.

We said last week that an evaluator who cannot compel anything is not an auditor. So here is the fair scorecard. On the plus side, it happened in six days, the access described is real access to how models are built rather than a look at the finished product, and it is not exclusive: Anthropic says it is also talking to METR, the non-profit research group, and others. On the other side, Anthropic states plainly that there are no standards yet for what an embedded evaluator may see or how it should report, and no settled way to pay for it. So, in its own words, Anthropic will fund Accenture's work directly. The company being inspected pays the inspector. Anthropic says long-term funding should come from pooled or government sources. It also says the evaluators do not reduce its own responsibility for safety. Accenture's shares rose about 8% after hours.

Shipping recognises this problem immediately, because it is ours. The owner pays the class society. The yard pays for the plan approval. The equipment maker pays for the type approval, as TS Cyanergy did this week. Nobody in shipping pretends that arrangement is pure. What we did instead was build layers around it over 150 years, so that one paid relationship is never the only check: a flag State that answers to a government, port State control that inspects without warning and cannot be paid, IACS auditing the societies themselves, and insurers who pay when everyone else was wrong and so have their own reasons to look.

The AI industry has one layer so far, and it is the paid one. That is not a scandal. It is day one. But it is a useful reminder when a maritime AI supplier tells you their product has been "independently validated." Ask who paid. Then look for a second check that nobody paid for.

This week had one. Rijkswaterstaat's check of Shipping Technology's 5% fuel saving is exactly that second layer: a public body, measuring on real vessels, with nothing to gain. It is also the rarest thing in this newsletter. In seven stories this week, it is the only number checked by someone with no stake in the answer.

Three questions for your next AI contract. Who has tested this product other than the supplier, and who paid for that test? Is there any check on it that nobody paid for, from a regulator, an insurer or a customer's own trial? And if a result is disputed, whose measurement wins: the vendor's dashboard or yours? Put the answer to the last one in writing before the trial starts, not after.

Disclosure: this newsletter is produced with the help of Claude, an AI model built by Anthropic, the company discussed in this piece. We have tried to hold it to the same standard as every other company in this issue. [Anthropic · CNBC · TechCrunch · AiatSea, 13 September]

📊 Why It Matters — Strategic Impact Table

🔭 On Our Radar

• 🗳️ Who Is Actually Paying for Shipping's AI Accounts? — carried for a sixth week: every adoption figure still counts organisations rather than people, our reader survey is open and collecting, we monitor whether any manager, union or class society publishes provisioning guidance instead of general policy language, and we will publish our own findings when we have enough responses, including if they prove us wrong.

• 🚢 Does the 10% Get the Same Check as the 5%? — Shipping Technology's first saving was validated by Rijkswaterstaat and its new one was not, we monitor whether the propulsion-control figure goes through the same public field test, and track whether any other inland or short-sea authority starts offering independent fuel-saving validation as a service.

• 🛰️ Will the Bridge Show When the Position Is False? — Galileo's authenticated signal arrives in 2027, we monitor which GNSS receiver and ECDIS makers announce OSNMA and E6 support, track whether IMO performance standards or flag States begin requiring authenticated positioning, and ask whether the warning reaches the officer on watch or stays in a log.

• 🇮🇳 Does the Live Proof of Concept Become Normal? — JNPA made bidders prove it on its own port first, we monitor whether other port authorities, charterers or ship managers publish live-trial requirements in their AI tenders, and track how long the JNPA twin takes to connect all five terminal operators.

• ⚙️ Does Anyone Publish Scope Before the Headline? — a non-essential approval was sold as a landmark validation, we monitor whether class societies begin publishing plain-language scope summaries with each approval, and track whether owners start asking for the certificate itself at the tender stage.

• ⚖️ Does the AI Industry Get a Second Layer of Inspection? — the first embedded evaluator is paid by the company it inspects, we monitor whether METR or any other non-profit signs an agreement, whether any government or pooled fund pays for evaluation, and whether any maritime AI supplier accepts an outside tester it does not pay.

• 🔌 Who Owns the Digital Layer Inside a Chinese-Built Ship? — China Merchants is building a vessel operating system as its yards fill with orders, we monitor which equipment makers sign up, whether owners can export their data or swap layers, and whether the system appears as standard supply in newbuilding contracts.

📅 Critical Maritime AI Research Areas for Managers

1. ⚖️ Who Pays for Independent Verification of Maritime AI: nearly every validation in this market is paid for by the supplier being validated, and public checks like Rijkswaterstaat's are rare. Research should map every maritime AI validation published in the last two years by who paid for it, and test whether supplier-funded and publicly funded results differ systematically.
2. 🛰️ Authenticated Positioning and the AI Tools Built on Position: voyage optimisation, emissions reporting, collision-avoidance support and risk screening all assume the position is true. Research should measure how each class of tool behaves under spoofing, and define what an AI system should do when it is told a position cannot be authenticated.
3. 🚢 Arrival-Time Sailing at Sea: inland vessels now accept a destination and an arrival time and control their own power to meet it. Research should examine what the same approach needs to work for coastal and deep-sea just-in-time arrival, including who sets the arrival time, how port changes are passed to the ship, and how the saving is shared between owner and charterer.
4. 🏗️ Terminal Productivity Gains and What Reaches the Ship: a 21% crane productivity gain is measured in moves per hour, not in ship time saved. Research should trace whether terminal optimisation actually shortens port stays and berth waiting for callers, and how much of the gain ends up with the carrier, the terminal or the cargo.
5. 🇮🇳 Live Proof of Concept as a Procurement Standard: a major port bought AI only after a live test, which almost no ship manager does. Research should design a standard trial protocol for maritime AI, covering baseline, duration, success measures, who measures, and how disputes are settled, that a mid-size manager could actually run.
6. 🔌 Data Ownership in Yard-Supplied Onboard Platforms: shipbuilding groups are starting to supply their own vessel operating systems. Research should compare newbuilding contract terms on onboard data ownership, storage location, export rights and third-party software access, and publish a model clause set for owners.

📈 Top Investment Opportunities

1. ⚖️ Independent Evaluation Paid by Someone Other Than the Supplier — the AI industry's first embedded evaluator is funded by the company it inspects, and maritime AI has almost no evaluation that the supplier did not pay for — the opportunity is in pooled or insurer-funded testing services, owner-side trial management, and measurement methods strong enough to settle disputes between vendor and customer — Anthropic + Accenture
2. 🛰️ Spoofing-Resistant Positioning on the Bridge — Galileo's authenticated signal becomes a free public service from 2027, but it only helps ships whose receivers and bridge systems can use it and show the result — the opportunity is in authenticated GNSS receivers, retrofit kits for existing bridges, integrity monitoring that feeds AI tools, and crew alerting that works at night — ESA · Galileo SAS
3. 🚢 Supervised Autonomy for Ordinary Vessels — a track pilot running on 200 inland vessels now controls propulsion too, with one fuel figure publicly validated, on conventional engines and gearboxes rather than showcase ships — the opportunity is in retrofit autonomy for the existing fleet, arrival-time control linked to port schedules, and the validation services that turn a vendor number into a public one — Shipping Technology
4. 🏗️ Optimisation Layers on Existing Terminal Equipment — 21% more crane moves per hour without adding a single crane, and a roll-out to three more terminals, shows the value sits in software over assets already paid for — the opportunity is in yard and equipment optimisation for mid-size terminals, and in the data links that let carriers see and plan around the gains — Yilport + Kaleris
5. 🇮🇳 Port-Wide Digital Twins Bought on Proof — a $9.7m contract won through a live proof of concept, joining five terminal operators, customs and rail into one operating picture — the opportunity is in integration across port stakeholders who do not share systems, proof-of-concept services as a paid product, and command-centre operation as a long-term contract — JNPA + SoftTech

📅 Top Monthly Picks

1. ✅ Three Certificates in Six Days, and Every One Stopped at the Same Place — Korean Register, Normec Verifavia and Bureau Veritas each certified maritime software and each excluded the result the buyer cares about; this week's ABS approval of a "non-essential" system sold as a landmark shows the pattern is still running, and the scope line is still the part nobody reads — Splash247
2. 🧭 The People Building AI Asked for the Brakes — four rival AI chiefs called for a slowdown in one news cycle and one of them promised outside inspectors; six days later the first one was named, and it is paid by the company it inspects, which is exactly the test we said we would apply — AiatSea, 13 September
3. 🌊 Koch Renews and Expands Sofar's Wayfinder — a renewal is the one piece of evidence that costs a customer money, and it remains the best buying signal in this market; read it beside this week's Yilport roll-out to three more terminals, which is the same signal from the port side — Splash247
4. 🛳️ CMA CGM Designs an AI-Assisted Containership and Asks How Much Control Stays With the Crew — the design study put the human question in writing before the steel was cut; this week's inland pilot answers it in daily service, with the captain setting the goal and approving the system's choices — Splash247
5. 🔍 Evergreen Asks ClassNK to Audit the Measurement, Not the Software — the first time a class society was asked to verify how a voyage-optimisation saving is measured; together with Rijkswaterstaat's check this week, it is still one of very few outside checks on a maritime AI number — Splash247

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