Last week the question was whether you can trust what the screen shows. This week the question is what the machine is allowed to do. An AI agent can now book freight at your company's rate without anyone logging in. Cameras on a bridge count near misses, and a ship manager has put the result in public. New tools read the whole company inbox and answer questions about it. And a survey says most people in shipping use AI every day, while fewer than one in ten work in a company that has rules for it. None of this is bad news. But the tools are moving from giving advice to taking action, and most companies have not yet written down who says yes. This issue is about that gap, and it ends with a short list you can copy.

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

🔗 Quick Links

• 🎥 Acheon Akti says camera AI cut bridge near misses almost 20 times — Splash247

  One pilot ship, the manager's own figures. Time outside procedure fell 69%. Nobody has said who counted the near misses, or for how long

• 🧮 63% use AI every day, only 8% say it is governed — Smart Maritime Network

  A Marcura and Thetius survey of 60 people. More than half spend an hour or more each week checking what the AI wrote. Small sample, and Marcura sells AI

• 🤖 Flexport lets outside AI agents book freight — Splash247

  Your own assistant can search rates, track cargo and book at your company's rate. Flexport says every exception goes to a human. No error rate published

• 🤝 Chartering chiefs ask who trains the juniors as Sea launches an AI assistant — Splash247

  Fednav, Vale and Western Bulk say everyone has the same data now. The edge is people. Sea's new assistant reads charter parties, with no customer or accuracy figure given

• 📨 Ulysses and GTMaritime both go after the inbox — Smart Maritime Network

  One tool searches all company email by voice or text, the other turns messages into tasks. No customer or result named by either. Our own team works in this market too

• 💰 PSA and Granite Asia launch a $50m port technology fund — Splash247

  The Netherlands also backs 13 maritime technology projects with more than €20m. A port operator and a government are paying for the next tools, not only vendors

• 📝 AiatSea: what may the agent do without asking? — AI at Sea

  72% of companies are using or planning AI that takes action. 15% say their rules are ready. Three short lists that close the gap

🚀 Big Moves This Week

1. A Ship Manager Publishes a Number for Camera AI, From One Ship

Ship manager Acheon Akti, which runs more than 65 vessels, has finished a pilot of an AI monitoring system from EVI Safety Technology. The system uses the cameras already on board. It watches chosen activities, flags possible risks to the crew, and keeps the footage for the office to review later. The video is processed on the ship, so it works when the satellite link is poor.

The numbers. Acheon Akti says bridge near misses on the pilot ship fell by almost 20 times. Other watched areas improved by at least three times. The total time the crew spent outside procedure fell by 69%, from 8,839 minutes to 2,730 minutes. The manager now plans to put the system on more ships, together with EVI's KARCO safety training.

Why we lead with it. In August a serving master mariner asked in print what these cameras really measure. We noted then that this kind of system runs on more than 1,300 ships and almost nobody had published a result. We have kept one question open since then: will anyone publish a number? This week a ship manager did. That deserves credit.

Now the honest part. It is one ship. The figures are the manager's own, and Splash247 says clearly they do not describe the fleet. We do not know how long the "before" and "after" periods were. We do not know who counted the near misses. If the camera counted them, then the tool is marking its own homework. And people behave differently when they know they are watched. That effect is real, but it can fade after a few months.

What to ask before you buy. Three things. Who counts the events, the system or a person? What happened to the numbers after six months, not six weeks? And did real incidents and claims go down, or only the things the camera can see? A drop in recorded deviations is a good sign. It is not yet proof of a safer ship. [Splash247 · AiatSea, 30 August]

2. Almost Everyone Uses AI, Almost Nobody Has Rules for It

Marcura published a report this week called Earning Trust: AI in Maritime. The consultancy Thetius wrote it, from a survey of 60 people working in the industry.

The headline numbers. 63% say they use AI every day. Only 8% describe their own company as mature and governed in how it uses AI. That is the gap in one line.

The checking cost. 55% spend at least one hour a week checking or fixing what the AI produced. Almost one in five spend five hours or more. The report calls this the "verification tax". Even so, 85% say AI saves them time overall. Seven in ten always check an AI recommendation before an important business decision. And 80% say a named person should stay responsible for a commercial decision, however good the AI becomes.

The part that matters most. 72% say their company already uses, tests or plans AI that can take action by itself, often called agentic AI. Only 15% say their rules are ready for it. Fewer than one in ten say their company's data is fully ready for AI to rely on.

Read the limits. Sixty people is a small sample, and we do not know how they were chosen. People who answer a survey about AI are more likely to be people who use AI. Marcura also sells AI products, and we covered its invoice-reading platform on 6 September. So treat the exact percentages with care. The direction is believable, because it matches what we see in every office: people started using the tools long before the company wrote a policy.

Why a ship manager should care. We have asked for eight weeks who is actually paying for the AI accounts your staff use. This report does not answer that. But it confirms the bigger point. Use is ahead of rules. If your company is in the 92%, the first step is small. Find out which tools your people already use, and what company data goes into them. [Smart Maritime Network · Splash247 · AiatSea, 6 September]

3. An AI Agent Can Now Book Freight, and the Limits Are Written Down

Flexport, the freight forwarder, has opened its platform to outside AI agents. A customer can connect Flexport to the AI assistant the team already uses, such as Claude, ChatGPT or Microsoft Copilot. Then a person can ask in plain words to track a shipment, list every delay or customs hold, search rates on a lane, or make a booking.

What is new. The agent does not only read. It can book. Flexport says an agent can send the cargo details, book at the company's own agreed rate, and get back a booking number and transit time, with no login to the Flexport platform. Rates cover ocean full and part container loads, and air freight.

The limits Flexport has published. The agent works inside each user's existing permissions, so it cannot do more than that person could do. An administrator must switch the connection on for the company first. The price, routing and cargo details are checked before a booking is made. Flexport also says every exception goes to a human expert, and every action can be traced.

What is missing. There is no error rate. We do not know how often an agent booked the wrong thing, or who pays when it does. Flexport calls this a first for the industry, and that is its own claim. And this is freight forwarding, not ship operation. No ship is being steered here.

Why it still matters to you. On 13 September we asked whether any maritime software company would publish limits on what an agent may do. This is the clearest example so far: act within the user's rights, check before booking, send problems to a person, keep a record. Use those four points as a test. When a supplier offers you an "agent" for purchasing, port calls or crewing, ask which of the four it has, in writing.

Disclosure: the connection standard Flexport uses was created by Anthropic, and Claude is one of the assistants named. This newsletter is produced with the help of Claude. [Splash247 · Flexport · AiatSea, 13 September]

4. Everyone Has the Same Data Now, So Who Trains the Juniors?

At the Splash Singapore conference, a panel of chartering chiefs talked about AI on the fixing desk. Chartering is the work of matching ships and cargoes and agreeing the price. In the same week, software company Sea launched Cora, an AI assistant that answers questions about charter parties and recaps inside Sea's contract platform, so a team does not have to search the files by hand.

What the panel said. Jesper Klarup of Fednav said competing desks use the same two or three platforms, and often price a cargo within a few cents of each other. His warning was short: the digital tool is necessary, but it can never become your alibi. Torbjørn Gjervik of Western Bulk said that if you only use the data in those systems, you will not beat the market. Michelle Gonzalez of Vale said data alone does not make better decisions, and that AI is still weak on context and politics.

The question nobody could answer. It came from the audience. If AI does more of the basic work that juniors used to learn from, who becomes the expert of tomorrow? Klarup said young people must still learn to run a calculation on paper. Willem Vermaat of Heidelberg Materials Trading said companies must let juniors make mistakes. Stephen Fletcher of AXSMarine said to get young staff away from screens and onto ships and into ports.

The other side. In this week's Marcura survey, 75% said AI helps less experienced colleagues become useful faster. Both things can be true. A junior with AI gets to the answer sooner, and may understand it less.

About Cora. Sea has not named a customer or given an accuracy figure. A wrong answer about a laytime or demurrage clause can cost real money, so the tool must show the exact clause it used. We covered Sea on 13 September, when it put BIMCO clauses into its workflow.

For ship managers. This is not only about chartering. It applies to your junior superintendents and purchasers too. Pick the tasks a trainee must do by hand at least once before the tool is allowed to do them. [Splash247 · panel · Splash247 · Cora]

5. Two Companies Go After the Inbox, and So Do We

Shipping runs on email. This week two software companies launched tools that try to bring order to it.

Ulysses Systems launched Finder. It reads the company's stream of emails and documents and sorts them by ship, port call, defect or other activity. A user can then ask a question by text or voice. Ulysses gives examples: why was a vessel delayed in port, what was done after a vetting observation, does the engineer's travel match the ship's latest arrival time? Ulysses says Finder can map information across more than 10,000 maritime processes and topics. The work builds on research with York University in Canada and IBM Canada.

GTMaritime launched GT Connect. It links a conversation to the vessel, the people, a task and the documents, so a discussion becomes a job with an owner. GTMaritime says it serves around 17,000 vessels and its mail system carries more than one million messages a day. In July it added AI that sorts and summarises incoming ship email.

The honest part. Neither company named a customer for the new product, and neither gave a result. No time saved, no accuracy figure. GTMaritime does not describe GT Connect itself as AI. And the Finder name is not new: Digital Ship reported an earlier Ulysses Finder in 2024, built at the request of the gas shipping company Exmar. So this is a new AI version of an existing idea, not a first.

The questions that matter. A tool that reads every email knows a lot. Who can search whose mailbox? Can a manager read a private message from a crew member? Does the answer show the original email, so you can check it? Where is the mail stored, and is it used to train anyone's model? Ask these before the demo, not after.

Disclosure: SeaQuest, the company behind this newsletter, builds BuzzRelay, an AI assistant for email, and fleet management software. Both companies in this story are competitors in some way. Read our view with that in mind. [Smart Maritime Network · Ulysses · Smart Maritime Network · GTMaritime · Splash247 · Digital Ship, 2024]

6. A Port Operator and a Government Put Money Behind the Next Tools

Two funding announcements came in the same week. Neither is about a product you can buy today. Both show who is paying for the next ones.

Singapore. Port operator PSA International and investment firm Granite Asia launched the G&P Strategic Innovation Fund, worth $50 million. It will invest in companies building technology for ports, logistics and supply chains, with a focus on AI, robotics and automation. Granite Asia, which describes itself as an $11 billion investment firm, will manage the fund and invest from its own funds alongside it. PSA says the fund supports its AI plans. PSA already has its own investment arm, PSA Ventures.

The Netherlands. According to Splash247, the Dutch are putting more than €20 million behind 13 maritime technology projects. They include AI for maintenance, robots for shipbuilding and autonomous craft. We have not yet seen the full project list, so we will not say more than that.

Why this is worth a few minutes. Most maritime AI is sold by a vendor to a customer who then has to work out whether it is any good. Here the customer side is paying to shape the tools. A port operator that invests in a startup can also test it on a real terminal. That is the same idea as India's JNPA, which made bidders prove their system on its own port before signing.

The honest part. A fund is a promise, not a result. No investment has been named. And $50 million is a modest sum in venture capital, split over many companies and several years.

For ship managers. If your ships call at PSA terminals, the tools tested there will reach you through berth planning and port data. For the Dutch projects, the one to watch is AI for maintenance. Ask who measures the result, and whether it will be published. [Splash247 · PSA · Container News · Splash247 · Netherlands]

7. AiatSea: What May the Agent Do Without Asking?

For two years, AI in shipping mostly gave advice. It wrote a draft, suggested a route, summarised a report. A person then decided. That is changing. This week an AI agent could book freight. In the Marcura survey, 72% said their company uses, tests or plans AI that takes action. Only 15% said their rules are ready.

Shipping already knows how to handle this. We do it for people every day. A purchaser can approve an order up to a set amount. Above that, the superintendent signs. Above that, the technical manager. A junior officer keeps the watch, and calls the master when the standing orders say so. Nobody finds this strange. It is simply written down.

Almost no company has written the same thing down for AI. So here is a start. It is three short lists.

List one: the agent may do this alone. Read, search, sort and summarise. Write a draft. Prepare a form. Remind a person. Nothing leaves the company and nothing is changed in any system.

List two: the agent may do this after a named person says yes. Send an email outside the company. Place an order or make a booking under a set amount. Change a record in the maintenance or crewing system. The person who says yes is responsible, the same as if they had done it by hand.

List three: the agent may never do this. Anything that changes how the ship is navigated or operated. Signing or accepting a contract. A payment above the limit. Sending crew personal or medical data to an outside tool.

One rule under all three. Every action leaves a record: what the agent did, what data it used, and who approved it. If you cannot find that record a month later, the tool is not ready.

This matches what people asked for this week. In the survey, 80% want a named person to stay responsible. On the chartering panel, Fednav's Jesper Klarup said the tool can never become your alibi. And Flexport published the same shape of limits for its booking agent.

What to do on Monday. Take one tool your office already uses. Write its three lists on one page. Show it to the people who use the tool, and ask what is missing. You will have a better AI policy in one hour than most companies have after a year of meetings.

Disclosure: this newsletter is produced with the help of Claude, an AI model made by Anthropic. It sits in list one. It reads and drafts. A person checks the sources and decides what is published. [Splash247 · survey · Smart Maritime Network · Flexport]

📊 Why It Matters — Strategic Impact Table

🔭 On Our Radar

• 🗳️ Who Is Actually Paying for Shipping's AI Accounts? — carried for an eighth week: this week's survey counts daily users but still not who pays or which accounts they use, 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 Camera Number Hold After Six Months? — Acheon Akti published a result from one pilot ship, we monitor whether it publishes figures from the wider roll-out, whether it says who counted the near misses, and whether any manager or insurer publishes incident or claims data next to the camera data.

• 🧮 Does Anyone Run a Bigger Survey? — the 63% and 8% figures come from 60 people, we monitor whether a class society, union or flag State publishes a larger sample, and whether any company publishes how it measures the time its staff spend checking AI output.

• 🤖 What Happens When the Agent Books the Wrong Thing? — Flexport published limits but no error rate, we monitor whether it or any customer reports a wrong booking and who paid for it, and whether any maritime software supplier publishes the same four limits for an agent used in purchasing, port calls or crewing.

• 🤝 Who Writes the Training Plan for the AI Age? — the chartering panel had no answer on how juniors learn, we monitor whether any owner, operator or manager publishes a training plan that names tasks to be done by hand first, and whether Sea names a customer or an accuracy figure for Cora.

• 📨 Who May Read Whose Email? — two inbox tools launched without a customer or a result, we monitor whether Ulysses or GTMaritime publish access rules, storage location and a named user, and whether any seafarer union raises the question of crew messages inside these tools.

• 💰 Where Does the Money Go First? — PSA's fund and the Dutch grants have named no results yet, we monitor the first investments from the G&P fund, the full list of the 13 Dutch projects, and whether any of them commit to publishing measured outcomes.

• 🛰️ Is a Second Position Source Becoming Normal? — following last week's spoofing stories, Splash247 reports Höegh Autoliners is installing PntGuard across its fleet and insurers DNK and NorthStandard are encouraging it, we monitor whether any of them publish detection results from ships in service.

📅 Critical Maritime AI Research Areas for Managers

1. 🎥 Do Camera AI Results Last?: the first published figure covers one ship over an unknown period, counted by an unknown method. Research should follow camera monitoring on several ships for at least a year, with events counted by people outside the supplier, and compare recorded deviations with real incidents and claims.
2. 🧮 The Real Cost of Checking AI Output: more than half of surveyed staff spend an hour or more each week checking what AI wrote. Research should measure this time by task type in ship management offices, and find which tasks save time after checking and which do not.
3. 🤖 Limits for AI That Takes Action: agents can now make bookings, and most companies have no written rules for them. Research should define a standard set of limits for maritime agents, covering user rights, checks before action, handover to a person and record keeping, and test it on real purchasing and port call work.
4. 🤝 How Juniors Learn When AI Does the Basic Work: senior chartering staff fear that trainees will reach answers without understanding them. Research should compare trainees who learn tasks by hand first with trainees who start with AI tools, and measure how each group handles a case where the tool is wrong.
5. 📨 Privacy and Access in Email AI: tools that read the whole company inbox raise questions of who can see what. Research should map the legal and practical limits for ship managers, including crew messages, different national privacy laws, and what a seafarer must be told.
6. 💰 Does Buyer Funding Produce Better Tools?: port operators and governments are now paying for development. Research should compare tools built with buyer money and live testing against tools built by vendors alone, on how often they are still in use two years later.

📈 Top Investment Opportunities

1. 🤖 Control Layers for AI Agents — agents are starting to book, order and send, and only 15% of surveyed staff say their company's rules are ready — the opportunity is in permission and approval layers that sit between any AI assistant and maritime systems, action logs that an auditor can read, and spending limits that match existing purchasing rules — Flexport
2. 🎥 Independent Measurement for Onboard Camera AI — a manager has published a first result from one ship, and buyers will now ask every supplier for the same — the opportunity is in outside counting and review of camera events, long-term studies for insurers, and services that link camera data to real incident and claims records — Acheon Akti + EVI
3. 🧮 Tools That Cut the Checking Time — staff spend an hour or more each week checking AI output, and trust grows when the tool shows what it used — the opportunity is in assistants that show the source for every answer, track their own accuracy over time, and learn from the user's corrections — Marcura + Thetius
4. 🤝 Training Built for the AI Age — senior staff do not know how juniors will learn when the tool does the basic work — the opportunity is in training that makes trainees do the task by hand first, practice cases where the AI is wrong on purpose, and short courses for chartering, purchasing and technical staff — Splash Singapore panel
5. 💰 Port Technology With a Test Site Attached — a port operator is putting $50m into startups it can also test on real terminals — the opportunity is in co-investing next to operators who bring a live site, and in companies whose product can show a measured result on a terminal within months — PSA + Granite Asia

📅 Top Monthly Picks

1. 📡 Four Tankers "in Lima": The Position Everyone Throws Away Is the Evidence — impossible ship positions are proof of spoofing, and most data services delete them; this week Splash247 reports a whole new market in second position sources, with Höegh Autoliners fitting one across its fleet — Splash247
2. 🎓 Should Every Officer Be Trained to Doubt the Machine? — a call to make AI literacy part of the STCW training rules; this week chartering chiefs asked the same thing for the office, and had no answer for how juniors will learn — Seatrade Maritime
3. 🎥 ShipIn Becomes a Public Benefit Corporation With a Duty to Report — the camera AI company took on a legal duty to publish; this week a ship manager went first and published a result from a different supplier, which is the number we had been waiting for — Smart Maritime Network
4. 🧭 A Hundred Ships of AI Watchkeeping, and the Part Nobody Expected — CMA CGM's head of ship management described a 100-ship roll-out with no fleet result attached; read it next to this week's one-ship figure, and ask both for the six-month number — AiatSea, 13 September
5. ⚖️ Who Pays the Inspector? — we argued the check worth trusting is one nobody paid the vendor for; this week's survey was paid for by a company that sells AI, and the camera result was counted by an unknown hand, so the question still stands — AiatSea, 20 September

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