The UK keeps a Master's overriding authority over uncrewed vessels but puts the person ashore. IMO's MASS Code binds nobody until 2032.
Mandated, not measured found a supervisory regime that had almost everything. Statutory backing, named officers, required documents, a body to investigate failures and publish findings. What it lacked was any reliable way to establish whether the supervision had actually happened.
Maritime has the opposite problem, and that is why it closes this act. Here the rules are arriving after the vessels. Uncrewed ships are already working in British waters, under guidance that describes itself as temporary, while the international code that will eventually govern them has been in effect for a matter of weeks and is not binding on anyone.
Stress test. Claim under test (Articles 3 and 5): managed autonomy requires permission, feedback, and a human with effective authority to intervene. Domain: maritime. The question: what happens when the governance regime is still being written?
Where is the Master?
For most of the history of seafaring, master of a vessel was one of the least ambiguous roles in any industry. There is a ship. Someone aboard commands her. That person is answerable for her navigation and the safety of everyone on her. Identity, location, control, authority and accountability all sat in one body, on one bridge, and nobody had to think about whether they could be separated because there was no way to separate them.
Now take the body off the ship.
The UK still requires a Master. Marine Guidance Note 703, published in October 2024, sets out that the owner, operator, charterer, or any other person shall not prevent or restrict the Master of a remotely operated unmanned vessel from taking or executing any decision which, in the Master's professional judgement, is necessary for safety of life at sea.
Read that again with the vessel in mind. It is the classic statement of a master's overriding authority, the provision that exists so that no commercial pressure can override a professional judgement about safety. It has been preserved intact and applied to a person who is not on the ship, cannot see it directly, and is connected to it by a data link.
Everything Article 5 said about permission and authority is contained in that one sentence, and so is the difficulty. The authority has been kept. The body it used to live in has gone ashore.
An architecture built in pieces
Around that sentence sits a structure assembled quickly and in fragments.
The Workboat Code Edition 3 came into force in December 2023, and its second annex applies to vessels with no persons on board that are operated from a location remote to the vessel. It defines a Remote Operation Centre, which may be a permanent shore-based location, a mobile one, or a crewed vessel. It defines Remote Operators as any person, including the Master, engaged in the remote operation of such a vessel.
Beneath that, a guidance note issued in December 2024 covers vessels between two and a half and four and a half metres, granting a general exemption on conditions, in force until the end of 2027. The vessels must be under constant supervision by a human able to take control at any time. A safety assurance assessment must be declared before operation. Speed is capped at six knots.
That first condition is worth pausing on. Constant supervision by a human able to take control at any time is not a paraphrase of this publication's framework. It is national guidance, issued eighteen months before any international code existed, arriving independently at the same requirement from the other direction.
And then the tell. MGN 703 describes itself as interim information, in place until a formal pathway for the training and certification of remote operators is established. A separate certification framework was published in 2025 as a pilot, a generic standard that may be considered for approval. The instruments that govern this are honest about being provisional. They say so on their own first pages.
Already at sea
On 19 September 2025, at around half past twelve in the afternoon, the crew transfer vessel Iceni Legend collided with the uncrewed surface vessel X-18 in the Greater Gabbard offshore wind farm, about thirty miles from Lowestoft. Iceni Legend, which carries maintenance engineers out to the turbines, was reported holed. The Lowestoft lifeboat was launched to escort her. Her pumps coped and she returned to port under her own power. Nobody was hurt.
X-18 is remotely piloted, and one of thirty-one such vessels operated by a single company.
The Marine Accident Investigation Branch is investigating and has published no findings. I am not going to speculate about what they will say, and this article does not need them to have said it. The paragraph is here for one reason. Whatever the regulations are anticipating, it is already happening.
The code that has just arrived
In May 2026, at its 111th session, the International Maritime Organization adopted the first International Code of Safety for Maritime Autonomous Surface Ships. It came into effect on 1 July 2026. It applies to cargo ships. It is goal-based, covering design, navigation, remote operations, connectivity, cybersecurity, machinery, fire protection, watchkeeping and search and rescue obligations, and it requires that autonomous and remotely operated vessels meet standards of safety, security and environmental protection equivalent to those expected of a conventional ship.
It is not mandatory.
Member states are invited to test its use and to share what they learn. The framework for that experience-building phase is due to be developed at the IMO's next session in December 2026. Work on a mandatory code is expected to begin in 2028, with adoption targeted by July 2030 and entry into force, through amendments to the international convention on safety of life at sea, on 1 January 2032.
Two details in the code matter more than the timetable.
A ship is only a MASS once it has completed the approval process and holds a valid MASS Safety Certificate. Enhanced automation alone does not make a vessel one, and vessels that qualify enjoy no privileges over conventional ships. Autonomy here is not a description of what a system can do. It is a status that has to be granted, and granting it changes what the vessel is permitted to be.
And the code recognises that the roles and responsibilities of the master, remote operators and onboard crew need to be regulated according to the degree of autonomy of the ship. Not the capability of the software. The allocation of authority among people, indexed to how much the machine does without them.
The order of operations
I want to be careful about what I am claiming here, because it would be easy to put a motive into the IMO's mouth that it has not stated.
What the organisation says is that the non-mandatory period gives member states the opportunity to test the code's use, and that the experience gathered will inform the mandatory version. It does not say the field is too immature to regulate. That is my reading, not their words.
But whatever the reason for any particular unresolved provision, the architecture is unambiguous about its sequence. Observation first. Binding obligation second. The evidence is to be collected before the rule is fixed, and there are six years set aside to collect it.
Set that against the four domains before this one.
Article 7 found feedback arriving after the recoverable window had closed. Article 9 found a command issued without any reliable measurement of whether it was followed, and then a stronger command issued while the measurement was still being consulted on. Here the ordering runs the other way. The measurement is being built first, deliberately, and the command is scheduled to follow it.
That is a closed loop, designed as one, in international law.
Why this is not applause
Having said that, I would not want to end an act on it, because three things are wrong with the loop and all of them are visible now.
The evidence comes from volunteers. A non-mandatory code generates experience only from operators who choose to work under it, and the operators who volunteer are the ones already confident in their safety case. The phase will measure the best-behaved part of the industry and then write rules for all of it. Selection bias is not a flaw in the design here so much as a property of it, and there is no obvious way around it that does not involve making the code mandatory before the evidence exists.
Equivalence is measured against an assumption that is being removed. Requiring an autonomous vessel to be as safe as a conventional one is reasonable and hard to argue with, but conventional safety standards were built around the presence of people aboard. Some of what makes a crewed ship safe is a crew noticing things nobody specified in advance. Equivalence to that is a harder target than it sounds.
And nobody has established how to investigate one of these when it fails. Marine accident investigation depends on interviewing the people who were there, reconstructing what they saw and understood and decided. Remove them and the method has to be rebuilt around data that the investigator did not specify and may not be able to obtain. The X-18 investigation will be among the first serious tests of that, which is the second reason this article does not interpret it. The whole experience-building phase depends on a feedback path that has not been demonstrated to work.
A loop whose sensor is untested is a loop on paper.
What Act II found
Five domains, none of them artificial intelligence, and each one testing a different property of managed autonomy against a real engineering environment.
Vehicles showed that supervision has to be architectural. A system that was not designed to be supervised cannot be supervised afterwards by adding a person to watch it.
Energy showed that feedback has to arrive while intervention still matters. A protection scheme that operates correctly after the recoverable state has passed has not supervised anything.
Industrial automation showed that permission has to be enforceable rather than expected. A control that depends on a person choosing to invoke it is not a control, however real the mechanism behind it.
Healthcare showed that accountability has to be measurable, not merely assigned. A duty that leaves no trace of whether it was discharged cannot be relied on, no matter how many names are attached to it.
And maritime shows that sometimes the governing system has to learn what should be governed before it can specify the rule, and that building the mechanism to learn is itself an engineering decision, with the same failure modes as any other sensor.
None of that came from thinking about artificial intelligence. It came from investigations into a robotic cell in Alabama, a grid collapse across the Iberian peninsula, a child on a cardiology ward, a driverless car in San Francisco, and a code adopted in London eight weeks ago.
The harder question
Maritime autonomy is a few weeks into a deliberate international experiment intended to discover what its binding rules should eventually say. It has defined roles. It has designated authority, preserved in writing, even where the person holding it is a hundred miles from the vessel. It has flag states, classification societies, certifying authorities and an accident investigation branch. It has a timetable for turning experience into regulation, and a named session in December at which the framework for gathering that experience will be agreed.
By the standards of this series, that is the immature case.
The harder question is what happens when systems acquire comparable authority somewhere with no equivalent machinery for learning what should constrain them.
Sources
MGN 703, Information concerning the training and competence of Remote Operators working with Remotely Operated Unmanned Vessels certified under the Workboat Code Edition 3, MCA, 30 October 2024
MGN 705 (M) Amendment 1, Remotely operated unmanned vessels (ROUVs) of 2.5 metres to less than 4.5 metres in length overall, MCA, 19 January 2026
The Workboat Code Edition 3, in force 13 December 2023, Annex 2
Remote Operator Certification Framework, MCA, June 2025
MAIB current investigations register: collision between Iceni Legend and X-18, Greater Gabbard offshore wind farm, 19 September 2025
IMO, International Code of Safety for Maritime Autonomous Surface Ships, resolution MSC.595(111), adopted 22 May 2026, in effect 1 July 2026
IMO press briefing and autonomous shipping FAQ, May and June 2026