Updating a Ship at Sea: Trusted Software Delivery and Knowing What’s Inside
Fifth of eight. Software always needs updating, and delivering it to ships that are often out of contact is harder than it sounds. Get the update channel wrong and you have handed over a master key.
Software is never finished. Bugs are found, improvements are made, and — crucially for security — new vulnerabilities are discovered that need patching. For an ordinary computer this is routine: it downloads an update over the internet and installs it. For an autonomous vessel, this everyday task becomes surprisingly hard, and getting it wrong opens one of the most dangerous doors of all.
Two problems at once
Delivering software to a fleet of vessels combines two awkward challenges. The first is connectivity. Vessels at sea often have slow, expensive or intermittent links, and some operate for long stretches with no connection at all. You can’t assume a vessel will simply reach out and download the latest version whenever you release it; it may not be reachable for days.
The second problem is trust, and it is the more serious of the two. A software update is, by its nature, new code that the vessel will run with full authority. If an attacker could slip their own code into that update channel, they wouldn’t need to fight through the vessel’s other defences — they’d simply have the vessel install their attack for them and run it willingly. The update mechanism is a master key, and it has to be protected accordingly.
Proving an update is genuine
The cornerstone of safe updates is the same idea we met with commands: cryptographic signing. Before any software is sent out, it is signed by MindChip using a key that only we hold. The vessel carries the matching public counterpart, which lets it verify that signature. When an update arrives, the vessel checks the signature before installing anything. If the software wasn’t signed by us — or if a single byte has been altered since we signed it — the signature won’t match, and the vessel refuses to install it.
The elegance of this approach is that it doesn’t depend on the delivery route being secure. The update could travel over an untrusted network, sit on a memory stick, or pass through several hands — it doesn’t matter. Tampering breaks the signature, and unsigned software is rejected. Trust is anchored in the cryptography, not in the courier.

Knowing exactly what’s inside
Modern software is assembled from many building blocks, often written by other people — open-source components, libraries, tools. This is normal and efficient, but it means a vulnerability discovered in some widely used component can suddenly affect thousands of products that quietly depend on it. Nobody sets out to ship a vulnerable component; they ship a component that was fine on Tuesday and had a serious flaw published on Wednesday.
To manage this, MindChip maintains what is essentially an ingredients list for every piece of software — a detailed inventory of every component and version it contains. In the industry this is called a software bill of materials, or SBOM. With that inventory in hand, when a new vulnerability is announced in some common component, we can answer a vital question in minutes rather than weeks: does this affect us, and which vessels? Before any software is packaged for delivery, it is also automatically scanned against databases of known vulnerabilities, so that problems are caught and fixed on our side long before the software ever reaches a vessel.
This is also, increasingly, not optional. The EU Cyber Resilience Act sets vulnerability-handling obligations on manufacturers of products with digital elements, and knowing your own software composition is the foundation of all of them.
Delivering to the reachable and the unreachable
Because vessels differ in how well-connected they are, safe delivery needs more than one route. A vessel with a good link can receive updates through the fleet’s command systems, brokered securely so that the vessel never needs to hold sensitive access credentials of its own. A vessel that is offline needs another path entirely: verified, signed software can be carried physically and applied during a port call, with the vessel performing all its integrity checks locally, without needing to phone home. In every case the vessel does the same fundamental thing before trusting new code — it verifies the signature. The route in changes; the discipline does not.
From the CertifAI project — making the ingredients list useful
CertifAI: a three-year EU-funded research project (2023–2026) on AI-assisted cybersecurity certification. MindChip led its maritime use case
A software bill of materials is only as valuable as what you can do with it. A raw inventory of a modern application can run to hundreds of components and tens of thousands of lines of machine-readable text — far past the point where anyone reads it.
MindChip supplied real bills of materials from the Artificial Captain platform to CertifAI tools designed to enrich them: correlating each component against vulnerability databases, exploit-likelihood scores and remediation history, then ranking what to fix first by a combination of impact and effort.
Our feedback to the project was blunt where it needed to be — an enrichment step that flags issues which don’t apply to the versions actually in use costs a security team more time than it saves, and precision matters more than volume. That is exactly the kind of finding a research project needs from a real user, and it is why use cases like ours were part of the project in the first place
Why this matters to you
For a customer or partner, the takeaway is reassuring. A vessel that only ever runs software cryptographically proven to come from MindChip, that is scanned for known weaknesses before it ships, and whose exact contents are documented and tracked, is a vessel whose behaviour you can reason about and trust. The update channel, which could so easily be the weakest link, becomes instead one of the strongest.
In the next post we look at something stronger still — what it means to prove a security property rather than test for it.
About this series. This post is part of MindChip’s “Cybersecurity for Maritime Autonomy” series, sharing practical insight from our work securing, assessing and certifying autonomous surface vessels. MindChip OÜ led Use Case 3 — the maritime use case — in CertifAI (certifai.info), a three-year Horizon Europe research project that developed AI-assisted tools for cybersecurity certification. MindChip was one of eleven partners, alongside Tecnalia, Hitachi Rail GTS Austria, Schneider Electric, TTTech, DNV, NTNU, Simula Research Laboratory, UBITECH, Catalink and EZU.

Funded by the European Union under Grant Agreement No 101120606. Views and opinions expressed are however those of the author only and do not necessarily reflect those of the European Union or the granting authority. Neither the European Union nor the granting authority can be held responsible for them.
