Avionics Software Loading: The Signature That Separates Maintenance From Compromise

A technician loads avionics software onto an aircraft component in a hangar, verifying the digital signature on a laptop screen.

A data load is one of the very few maintenance actions that leaves no visible trace on the aircraft. Nothing is removed, nothing is torqued, nothing changes weight. A laptop is connected, a part number is selected, a progress bar runs, and the aircraft is in a different configuration from the one it was in ten minutes earlier. Whether that difference is an approved modification or a compromise depends on a verification the hangar environment does very little to force anyone to perform. That is why software loading is where continuing airworthiness and information security stopped being separate subjects.

What is actually being loaded

The unit of work is not a file. It is a loadable software aircraft part, and the word « part » is doing real work there. It carries a part number, appears on an approved configuration list, its installation is a maintenance action, and replacing it changes the configuration in which the aircraft is approved to operate. The industry format for such a part — a media set with a header file, a list of the data files it contains and integrity values over each of them — comes from the ARINC 665 family.

Around it sit the transport standards. ARINC 615 covers the older data loading over the aircraft’s own buses; ARINC 615A covers loading over Ethernet, from a portable data loader or from a laptop running loader software, which is what most line and shop work uses now. ARINC 827 defines the container used to move a software part electronically from supplier to operator — the crate — and it exists precisely because the physical medium that used to carry the part, and to carry implicit custody with it, went away.

That disappearance is the whole story. When a software part arrived as a numbered, controlled physical item shipped by the type certificate holder, custody was a physical problem with a physical answer: a locked cabinet, a goods-in register, a signature on a delivery note. When the same part arrives as a file, over a network, into a software library, and then onto a laptop that also reads email, the only thing still connecting what is in the loader to what the manufacturer approved is cryptography.

A software part is like a digital component of the aircraft. It’s not just a file—it’s a certified piece that changes how the plane works. Before, it came on a physical disk with a paper trail. Now, it’s sent electronically, and the only way to trust it is through digital signatures.

An A320neo receives a new navigation database via a laptop. The loader checks the digital signature to ensure it’s from Airbus, not a hacker. Without this check, a malicious file could alter flight paths.

The CRC answers a different question

Loads carry integrity values: a CRC, sometimes a hash, computed over each file and over the media set as a whole. Loaders check them, targets check them, a mismatch aborts the load. It is easy, and common, to treat that check as the security control. It is not one.

A technician in a dark uniform uses a laptop connected to avionics equipment in a hangar.
Secure software loading prevents unauthorized access to critical systems.

A CRC is engineered to catch the failure modes of a transmission channel: a flipped bit on a bus, a truncated transfer, a medium that degraded in a toolbox over a summer. It does that job cheaply and very well. What it cannot do is resist somebody who wants the altered file to pass, because the algorithm is public and the value is recomputed from the content. Change the content, recompute the value, and the load is internally consistent again. A cryptographic hash raises the cost of forging a matching file, but inherits the same structural weakness whenever the hash travels alongside the file it describes: whoever can alter one can alter the other.

A digital signature is a different kind of statement. The publisher computes a hash of the part and transforms it with a private key that never leaves the publisher. Anyone holding the matching public key can then establish two things in a single operation: that the content has not changed since it was signed, and that it was signed by the holder of that private key. The first is integrity, which the CRC also gives. The second is authenticity, which the CRC cannot give at any price and never claimed to.

What each mechanism actually proves about a load
Mechanism What it establishes What it leaves open
CRC over files and media set The bits that arrived are the bits that were sent, within the detection power of the polynomial Who produced them, and whether anyone changed both the content and the value on purpose
Cryptographic hash published with the load The content matches the digest that accompanies it Whether the digest itself is the one the publisher produced
Digital signature over the part Content unchanged since signing, and signed by the holder of a specific private key Whether that key holder is entitled to publish for this aircraft, and whether the part is approved for this configuration
Approved configuration list and release to service That this part number is the one this aircraft is supposed to be running Nothing about the bits, which is why it does not replace the signature check

The last row is the one that gets skipped in discussion and matters in practice. Cryptography answers « is this what the publisher made »; the configuration record answers « is this what this aircraft should have ». Neither answers the other’s question, and a valid signature on the wrong standard is still a configuration error.

Traced end to end, a signed load passes through five custodies, and each one either preserves the signature or breaks it.

  • The publisher. The type certificate holder or the equipment supplier builds the part, signs it with a private key held in its own infrastructure, and issues it with a part number and an instruction saying which aircraft it applies to.
  • Distribution. The signature survives copying, which is exactly the property that makes electronic distribution acceptable at all: the file can be duplicated freely and remain provably the publisher’s.
  • The software library. The part is received, verified, catalogued against the fleet configuration, and held. This is where an organisation decides whether it has a controlled repository or a shared drive.
  • The loader. A portable data loader or a laptop pushes the part to the aircraft. It is the element that most often falls outside the security posture applied to the rest of the estate, because it is treated as a tool rather than as a computer.
  • The target. The receiving equipment writes the part into its own memory. Whether it verifies a signature before doing so is the most consequential design decision in the chain, and it is not the operator’s to make.

The industry has written this down. ATA Spec 42 frames the aviation public key infrastructure — who may issue certificates, under what policy, and how identities bind to organisations — because a signature is worthless without an agreed answer to « whose key is this ». ARINC 835 gives guidance on securing loadable software parts with digital signatures. On the airworthiness side, the DO-326A and ED-202A airworthiness security process specification, with its companion documents, brings security inside the certification argument instead of leaving it an IT matter running alongside.

Where the verification actually runs

Ask where a load is verified and the usual answer is « the system checks it ». Ask which system, and the answer separates fleets.

Verification points and what each one can and cannot catch
Point Catches Misses
On receipt into the software library A part that never came from the publisher, and a corrupted download Anything that happens to the copy afterwards, on the way to the aircraft
In the loader, before transmission Substitution inside the operator’s own environment, if the loader has a genuine trust anchor A compromised loader, which will report success whatever it sends
In the target, at load time Everything upstream, including a compromised loader, because the check is made by the equipment that will run the code Approval and applicability: a genuine part for the wrong standard still verifies
Post-load configuration report Wrong part number, partial load, an item that did not take Anything about the authenticity of what did load

The row that decides the shop’s role is the third one. Where the target verifies the signature itself, the loader and the library are conveniences: they make errors less likely, but the aircraft is defended by the aircraft. Where the target only checks a CRC — which is the case for much of the equipment certified before signed loading was an expectation, and which will remain the case for as long as those units stay in service — the last enforcement point in the entire chain is a procedure carried out by a person in a hangar. Any mixed fleet, and most fleets are mixed, contains both situations at once, sometimes on the same aircraft.

The trap. « The load completed successfully » is a statement about the transfer, not about the origin of what was transferred. On equipment that verifies only integrity, a successful load and a hostile load are indistinguishable on the loader’s screen. Before assuming a check exists, find out — per target, not per fleet — what the receiving equipment actually validates.

Certificates, trust anchors, and a hangar with no network

A signature is checked with a public key, and a public key means something only if its provenance is established. In practice that means a certificate chain ending at a trust anchor provisioned into the verifying equipment, and it is here that aviation’s timescales collide with normal cryptographic practice.

Ordinary information systems manage keys with short certificate lifetimes, frequent rotation and online revocation checks. None of the three transfers well to an aircraft. An airframe serves for decades, and equipment verifying a signature late in that life must still trust a chain provisioned long before. A parked aircraft has no route to a revocation service, so validation has to work entirely offline. And many avionics targets have no trustworthy wall-clock time, which makes naive expiry checking either impossible or dangerous: refusing a valid safety-related load because a wrong clock made a certificate look expired trades one failure mode for a worse one.

The consequences are practical. Key rollover has to be planned and provisioned as a maintenance event. Certificate expiry in the distribution chain produces parts that were validly signed but now fail verification, and the pressure that puts on a night shift with a departure to make is exactly the pressure that produces an override. Any procedure allowing verification to be skipped « because the certificate expired » has turned the whole apparatus into decoration.

The right reflex. Treat a verification failure as a technical event that stops the job and gets recorded, not an obstacle to work around. A failed check has two possible explanations — a real problem with the part, or a real problem with the process — and both need to be known.

Loading software is maintenance, and the record has to show it

Under the European system, an organisation loading software on a commercial air transport aircraft works under a Part-145 approval, the configuration is managed under the continuing airworthiness rules of Part-M with a Part-CAMO organisation owning the aircraft’s status, and the action ends in a release to service like any other task. The FAA framework reaches the same result through a different structure. Common to both: the aircraft is airworthy in a defined configuration, software part numbers are part of it, and an undocumented load leaves a state nobody can attest to.

Two developments pulled security into that same frame rather than leaving it beside it. Aircraft with networked architectures have for years been certified against special conditions addressing the security of those networks, which makes protection of the loading path part of the certification basis rather than a matter of good practice. And EASA’s Part-IS requires organisations across the European system, maintenance organisations included, to run an information security management system covering risks with a potential impact on aviation safety. An organisation that loads software is squarely inside that scope.

What actually goes wrong

Almost nothing that goes wrong in this domain is a broken algorithm. The recurring failures are mundane, and every one of them is a process the shop controls.

  • The right part on the wrong standard. A genuine, correctly signed load applied to an aircraft whose configuration does not call for it. Nothing in the cryptography objects.
  • Parts entering outside the library. A field representative arrives with a memory stick, the load works, the aircraft departs, and the controlled repository never saw the part.
  • The loader as an unmanaged computer. A general-purpose laptop, shared between technicians, connected to hotel networks, running loader software not updated in years, holding credentials for the aircraft-side connection.
  • Partial and interrupted loads. Aborted transfers leaving a target in an indeterminate state, cleared by a retry nobody records, so the configuration report becomes the only evidence of what happened.
  • No baseline to compare against. Detecting an unexpected software part requires knowing what the aircraft should be carrying. Where the approved configuration list is not readily available in the hangar, the post-load report cannot be read as anything but a receipt.
Shop-level controls: the question each one answers
Control Question it answers Evidence it leaves
Single controlled software library, with verification on receipt Did this part come from the publisher? A dated entry tying a part number to a verified signature
Managed, dedicated loader devices Is the machine that talks to the aircraft under configuration control? Asset record, software baseline, patch status of the loader itself
Applicability check against the aircraft configuration list Should this tail have this part number? The work card, signed, referencing both the part number and the aircraft standard
Post-load configuration report compared to the expected baseline Is the aircraft now in the configuration we intended? A printed or archived report attached to the task record
Verification failures logged, never silently retried Is anything happening we do not understand? An occurrence record, the only way a pattern becomes visible

Nobody decided to make maintenance organisations part of the aviation security perimeter. It happened structurally, when distribution went electronic and the physical custody that used to carry the guarantee stopped existing. The guarantee moved into a signature, and a signature only protects what somebody checks. On modern targets the aircraft makes that check, and the shop’s job is to avoid configuration errors. On everything else the check is made, or not made, by a technician following a procedure — and the shop is not part of the chain of trust, it is the end of it.

That is a defensible position provided it is a known one. The failure case is the organisation that assumes the aircraft is checking while the aircraft assumes the organisation is. Software libraries and loader control come up regularly in the maintenance discussions on the FL330 forums, always from the same starting question: which of our targets actually verifies a signature, and how would we know.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

FL330, c’est toute l’aviation professionnelle — explorez nos autres espaces.