Inside Trust-Buster: Building a Platform that Meets Legal and Security Challenges
· 8 min read · Updated
Lire en français
How we actually decide, including the decision we had to reverse
The Ops-Devices platform was designed to address technical needs and the legal and security requirements arriving with NIS2, the CRA and the Machinery Regulation, and that much is on the front page of every vendor in this space, so it tells you very little.
What follows is the part that is harder to write, because it covers how the legal and security decisions get made, what we refuse to claim, and the architectural commitment we had to unwind when the ground moved under it.
We keep a register of the alternatives we rejected
Most engineering teams document what they built, whereas we also document what we did not build, and why, a practice that is now common and for which resources such as the Architecture Decision Records are worth reading and using.
Every significant architectural choice goes into a register that records four things, the option retained, the options discarded, the reasoning at the time, and the part that matters most to us, the re-evaluation trigger. That last field is a written statement of what would have to change in the world for the decision to be reopened.
This started as ordinary R&D discipline, and it turned out to have two other uses we did not anticipate.
The first is regulatory, because both the CRA and the Machinery Regulation ask manufacturers to demonstrate that security was considered by design, not asserted afterwards, and a dated register of rejected alternatives with reasoning attached is a far better answer to that legal and security question than a policy document written the month before the audit.
The second is honesty enforcement, because it is uncomfortable to write down a re-evaluation trigger, since you are committing in advance to a condition under which you will have to admit that a decision no longer holds, which brings us to the one that fired.
The decision we had to reverse on our hardware trust chain
Our early hardware architecture was anchored on a European trust chain, because the cellular platform we had selected combined a microcontroller with a modem carrying a secure enclave whose IP, meaning the licensed design of the protected area of the chip where the keys live, came from Tiempo Secure, a Grenoble-area company founded in 2007, whose TESIC RISC-V secure element had passed Common Criteria EAL5+ evaluation with the French ITSEF laboratory SERMA, and which had been selected for a French State-backed sovereign post-quantum security microprocessor programme. The modem line came from Sequans, a French supplier of cellular semiconductors, so European board, European modem, European enclave IP, a chain we could describe end to end and defend in front of a defence procurement officer.
Then both pieces changed hands.
In August 2024 Qualcomm agreed to acquire Sequans' 4G IoT technology assets, including employees and licences and the Monarch product line, and the transaction completed the following month for around $200 million, with Sequans retaining a perpetual licence to continue using and commercialising the technology. Tiempo Secure was subsequently acquired by Qualcomm as well, a transaction confirmed publicly by its co-founder rather than announced by press release, and now reflected on Tiempo's own website.
What this did and did not change. It is worth being precise, because the loose version of this story is more dramatic than the true one. The technology did not stop working, Sequans continues to operate and holds a perpetual licence, Tiempo's engineering team remains in Montbonnot, and nobody's chip turned into a backdoor overnight.
What changed is narrower and, for us, decisive, because the sentence we had been writing was no longer true. We had described a fully European hardware trust chain, and after the acquisitions the enclave IP and a substantial part of the modem technology sat under US corporate ownership. For an industrial customer, that is a footnote, but for a defence customer evaluating supply-chain provenance and the digital sovereignty of its suppliers, it is the whole question.
What we did. We deprecated that platform as our sovereignty anchor and moved the anchor elsewhere, to an STM32 microcontroller, a Nordic cellular bearer, and a discrete secure element sourced separately rather than inherited through a modem. The prototyping path now runs on a Nordic development platform with an STSAFE secure element, with a higher-assurance discrete element on the production path.
What we learned, and this is the part worth generalising, fits in one sentence, sovereignty is not a property of a component, it is a property of a claim you have to keep true. Ownership of semiconductor IP changes hands on a timescale much shorter than the service life of an industrial machine, so any architecture that stakes its differentiation on the nationality of a supplier needs an explicit re-evaluation trigger attached to it, because that trigger will eventually fire, and the alternative to noticing it yourself is a customer noticing it for you.
Our register now carries that trigger on every sourcing decision that carries a provenance claim.
Three legal and security claims we refuse to make
Compliance-adjacent marketing has a gravitational pull toward overstatement, and these are the three places we have decided in advance not to follow it, written down so that a reader can hold us to them.
EAL5+ is inherited, not obtained. When we reference a Common Criteria EAL5+ evaluation, it belongs to a secure element component and to the laboratory that evaluated it, because Ops-Devices holds no such certification and does not claim one. A product containing a certified component is not a certified product, and the distinction matters precisely to the customers who care about certification at all.
Radio signal strength is a gate, not a proof. Our system uses received signal strength as an entry condition, so below a threshold the exchange does not begin, but it is not an authentication factor, because signal strength can be amplified, relayed and spoofed, and no amount of filtering turns it into a distance measurement. Time-of-flight ranging over ultra-wideband does that properly, but we do not have it today, and we say so rather than implying that careful thresholding is equivalent.
We cannot claim exhaustive detection. No radio-frequency monitoring system can demonstrate the absence of a signal it did not see, so detection is a corroborating input to a decision, never the anchor of one, and any vendor claiming complete detection coverage in an uncontrolled radio environment is describing an ambition, not a measurement.
Each of these makes a slide slightly less impressive, but each of them also survives contact with a customer's technical evaluation, which the impressive version does not.
Collaboration and innovation
Our development process does involve continuous engagement with the outside world, but it is worth describing what that actually consists of, because the generic version of this sentence claims more than we can support.
We design against published standards, and we do not sit on the bodies that write them. IEC 62443 shapes our architecture, the CRA and the Machinery Regulation shape our documentation, and IEEE 802.1AR, ISO 15118 and the IETF's constrained-environment work inform specific design choices. We are readers of those standards and, where relevant, we prepare structured responses for procurement teams who cite them, but we do not hold seats on standards committees or regulatory bodies, and describing our work as collaboration with regulators would be an overstatement of a different kind of relationship.
We build on open source, deliberately and by default. Firmware on open frameworks, published schematics in an open EDA format, off-the-shelf modules with documented protocols rather than bespoke hardware wherever a standard part will do, and this is not ideology, because it is the only way a small team keeps a supply chain it can actually audit, and it is what makes the provenance question above answerable at all.
The innovation is confined to one layer. Assembling proven components and open frameworks is not innovation, and we do not present it as such. What we keep in-house is the chain of trust that ties those components together, meaning local policy decision without connectivity, ephemeral signed capabilities, an evidence trail designed for retrospective proof rather than real-time detection, and enforcement that sits at the boundary of a machine rather than inside it. That is a narrow layer, and as far as we know it is also the layer nobody else in the competitive landscape occupies, which is why it is the only part worth building ourselves.
What the platform answers, and what it does not
A platform that claims to deliver compliance is describing something that does not exist, because compliance is an organisational outcome, and a product can only close specific legal and security requirements within it.
The Ops-Devices platform addresses a defined column of that matrix, access control and asset management for equipment that has no digital path, evidence of physical intervention that survives an offline site and an eighteen-month gap, and enforcement that does not require modifying the machine.
It does not address network segmentation, vulnerability handling, backup and recovery, cryptography policy across your estate, supplier due diligence, staff training, or incident response process. Those are real obligations under NIS2 Article 21(2), and anyone telling you a single product covers them is selling you a compliance narrative rather than a control.
The reason we start where we start is not that our column is the most important one, it is that on almost every site we visit, it is the one that is completely empty, with no list of who may operate the asset, no record of who did, and no mechanism that could produce either.
Why we write any of this down
There is a straightforward commercial argument against publishing a reversed architectural decision, an inherited certification caveat and a list of things a product does not do, because it makes the story less clean.
We do it because of who evaluates this category, since the people who buy access control for industrial and defence sites are, by profession, people who look for the gap between the claim and the artefact, and a vendor who names their own limits first is not being modest, they are removing the most productive line of questioning from the evaluation and leaving only the substance.
And we do it because of the register, because once you have written down the condition under which a decision stops being true, you have to do something when it happens. That is an uncomfortable discipline, and it is the closest thing we have to a guarantee that what we tell you next year on legal and security matters will still be defensible.
