AiatSea Analysis: Class Quietly Became the Auditor of AI Claims

One classification society appears three times in this issue, in three entirely unrelated stories, and never once as the technology. ClassNK verifies the methodology behind Evergreen's fuel-saving trial and issues a Statement of Fact on the result. ClassNK is the independent verifier for a Ro-Ro emissions data framework covering 80% of global car carrier capacity. ClassNK holds the Approval in Principle on the autonomy system going onto a limestone carrier in December, with a coastal notation as the target. That is not a coincidence, and it is not a story about one society. It is the shape of a role the industry has been quietly assigning. What changed. For three years the maritime AI conversation has been about capability — what the model can do, how many vessels it runs on, what percentage it saves. Capability claims are cheap to make and expensive to check, which is why the vessel counts kept climbing while published outcomes did not. The bottleneck was never intelligence. It was verification. Classification societies happen to hold the two things verification requires and almost nobody else in this market has: a technical competence to examine a methodology, and a reputational asset that would be damaged by signing off something false. What is genuinely new is the object being verified. Last week ClassNK certified an AI training platform — that is verifying a product, and the society published exactly which five capabilities it had examined. This week it is being asked to verify a measurement: not does the algorithm work, but is the number it produced arrived at honestly. Those are different jobs, and the second one is the one buyers actually need. Why this should make you slightly uncomfortable, as it does us. Verification is becoming a service, sold by the same organisations that also sell advisory, training and software. ClassNK contributed to the Ro-Ro framework and then verifies against it. That arrangement is completely normal in maritime standards work, it is how most industrial standards get written, and it is still a structure where the auditor and the architect are the same institution. We are not alleging anything; we are saying the question is legitimate and nobody in the industry is asking it out loud. The second discomfort is scope creep in the other direction. A Statement of Fact means a stated methodology was followed and a stated result obtained. It does not mean the result will hold on your ship, in your trade, with your hull condition. Vendors will quote it as though it does. We give it a year before "class-verified savings" appears in marketing where the class involvement covered a single vessel on a single trade for a single season. What to do now — the transferable part. When a vendor claims verification, ask three questions. What object was verified — the software, the methodology, or the result? Who paid for the verification, and could the verifier have said no? And what is the scope statement — which vessels, which conditions, which period? Get all three in the scope document, not the brochure. Last week we published ten questions we thought belonged in every maritime AI contract. This week the industry added the eleventh for us, and it is the sharpest one: who checked, and were they paid by the person making the claim? The honest caveat, applied to ourselves: three appearances by one society in one week is a pattern we have noticed, not a trend we have measured. We have not surveyed how often class societies verify AI claims, whether other societies are doing the same, or whether verification requests are actually increasing. This is editorial judgement about a direction of travel, offered as such, and if a reader has data that contradicts it we would genuinely like to see it. [AiatSea, 16 August · Evergreen trial · AiatSea reader survey]

This story is part of the Maritime AI Digest of 23 August 2026.