A future for Self Sovereign Identity?
Many children in Europe grew up with the tales of Baron Münchhausen, who claims to have lifted himself and his horse out of a mire by pulling his own hair. The image is so powerful because the problem of the circular dependency is so clearly visible. In real life, circular dependencies are often far less obvious.
Which is why the first article in this series was primarily focused on looking behind the SSI smoke and mirrors around Web based identifiers and communication protocols. The resulting discussions in the Rebooting the Web Of Trust (RWOT) community were quite enlightening, and included a deeper look at the EU Digital Identity Wallet Technical specifications.
One of the mirrors basically broke when claims of OpenID4VC supporting decentralized identifiers were shattered when someone pointed out that while the EU Wallet is marketed on digital sovereignty and privacy, but in reality does not does not allow decentralized identifiers:
So while it was clear that OpenID4VC and did:web* do not qualify as decentralized, Self-Sovereign Identity, some people advocated to just embrace the false marketing in the hope that it would create wider acceptance and the appearance of adoption for SSI.
But has that approach ever really worked?
More often this kind of “sovereignwashing” appears to run a high risk of creating false expectations, disappointment. Which would ultimately cement the status quo of the federated platform identity lock-in for the next 20 years. As a community we should focus on building actual decentralized identifiers, communication protocols, and applications.
Because the true social and economic value of SSI is not just in the identity layer itself, it is in the decentralized applications enabled as a result.
Some of which would be in direct competition to the champions of the platform age, who are investing their financial and political capital into OpenID4VC and Web based identifiers to prevent that competition from ever getting off the ground. A classic “old industry vs new technologies” battle.
There are real opportunity costs across most of economy and society if the old encumbents manage to postpone or kill innovation.
Security and privacy for a globally networked society
Technology and corresponding security have been head to head in a special kind of race for a long time, dating back to an Egyptian inscription around 1900 BC in the main chamber of the tomb of Khnumhotep II, over Julius Caesar using a ROT-3 cypher in 100 BC, all the way to the famous Enigma machine used in World War II. The more people potentially had access to a message, the harder the encryption had to become.
The encryption used by Julius Caesar was not particularly strong, because it relied on a supposedly secret algorithm. Once parties know the secret, encryption and decryption become trivial. Over time this moved to well-known algorithms using shared secrets. And even though the shared secrets are more complex on today’s internet, this fundamental principle hasn’t changed:
If you know the shared secret, and can intercept the encrypted message, you will be able to read, and also impersonate and falsify communication.
In contrast, Engima was quite strong for its day because it combined a rotating cypher with a codebook that was famously carried by U-Boats allowing them to choose the correct settings. Literally handed over to the commander of the boat by hand in a secure location before departure, these code books effectively represented a cryptographic key, shared over a second channel — the physical handover.
Which makes any well-designed encryption system almost impossible to break. Unless, of course, you have intimate knowledge of the inner workings of the rotating cypher, and can guess certain messages, like weather reports, to then use brute force to arrive back at the settings for the day. Those settings then allowed to read other messages, which would otherwise have been unbreakable.
In other words: The cryptography of the Enigma machine itself was solid, and essentially unbroken. But the Allied Forces were able to exploit structural weaknesses designed into the operation of Engima to attack the key generation for the day.
Security in Swiss Healthcare
That particular race accelerated when the Internet was born. In 1996, when the internet was still young, the US Congress deliberated and passed the Health Insurance Portability and Accountability Act (HIPAA). That same year, the Swiss Medical Association (FMH), realized patient data had to be better secured on the internet, leading to the creation of Health Info Net (HIN). Starting from encrypted email, Swiss doctors have relied on HIN for decades to keep their patient data safe.
But technology years are a lot like dog years. And 28 years is a very long time.
HIN is constantly working to innovate and improve its solutions. Which is how Vereign, working closely with our partner More than Bits, started to run some POCs with HIN in 2023, and ended up working all of 2024 almost exclusively for the Swiss healthcare sector.
Our challenge: Design a system that starts from what today’s users are used to, while re-thinking the system architecture using SSI and modern data ecosystem architectures, based on the work we had done for Gaia-X.
The starting point was obvious: Email is the world’s largest distributed identity database and communication protocol. It is the use case with which HIN started, and it is the singular product that all users rely on mutliple times each day to communicate with colleagues, laboratories, and patients.
Email is also facing challenges of concentration and capture by the large, federated platforms. And its lack of an identity layer has made it a fertile ground for attacks by malicious governments, corporations, and common criminals.
Vereign showcased its first prototype to harden email using SSI in 2019, which earned us a nomination as the hottest new innovation for the Swiss Digital Economy Award in Zurich. COVID-19 had other plans, but our experience proved invaluable when working on the POCs with HIN.
This time, we built out peer to peer email exchange via DIDComm. Secure, encrypted, authentic and designed in a way that it can be plugged into any legacy email system to gradually switch to a new, identity verified transport layer reaching all the way to the people themselves.
From prototyping to production: Quest for the identifier
We built these prototypes using did:web, because it is a great placeholder to stand in for decentralized identifiers while rapidly prototyping around user flow and experience.
But from the onset it was clear that did:web would not be the choice for production, because for all the reasons also highlighted in the last article:
Web based identifiers must never be used for personal identity.
Our preferred choice would have been KERI due to its robust security and privacy architecture. But with the official implementation being Python only, we had concerns about efforts required in supporting a secure, long term solution across the range of platforms we anticipated.
The Rust implementation by the Human Colossus Foundation fared better on that front. But there seems to be a rift in the community, causing concerns of diverging implementations, as well as long-term support. Which are exacerbated by the choice for European Public License (EUPL).
We could not find information about adoption, nor community. And finally, the security of KERI as a concept critically depends on the networks of Witnesses and Watchers, for which we could not find information about size, health and long term viability of these networks for either implementation.
Had we chosen KERI in February 2024, we would not have been able to go productive before these issues had been resolved. And our time line dictated we had to be ready for initial production by late 2024. As a result, KERI was a non-starter.
Other methods, such as did:indy, have been in decline for some time, and Sovrin is shutting down in just a couple of weeks. Methods like did:peer on the other hand are not great in scenarios where long-lived connections are desirable.
So in the end, our search for production ready decentralized identifiers that could safely be used for natural persons left us empty handed.
Ignoring XKCD
The competing standards comic by XKCD is a classic. As far as rules go, it is a good one. But there are no rules without exceptions. Having exhausted every other path, we decided to ignore XKCDs’ best practice. Only, we did not aim to create the universal solution — that’s KERI — but to create the simplest possible, yet still sufficiently safe identifier for the requirements of our specific use case.
Like any good design, it should build on existing technologies as much as possible, be simple enough to be implemented within a reasonable time frame, and to be supportable for at least 5–10 years, when potentially it would be replaced by something better.
Designing a decentralized identifier
Our requirements asked for an identifier that was truly secure and private. We explicitly sought to minimize dependencies on infrastructure such as DNS, Web Servers and Certificate Authorities. Blockchain would have fit these criteria, but we do not require a global consensus. All we needed was a decentralized storage system that would guarantee integrity and availability of records.
Git might have been an option. It is Content-Addressable Storage, so objects are referenced by their hash, any modification creates a new object. But Git would add unnecessary overhead, and there is a central repository. The Interplanetary File System (IPFS) on the other hand is built for peer to peer distribution between nodes without a central server.
Like Git, IPFS is built on Content-Addressable Storage (CAS). Objects are referenced by their sha256 hashes. Users can request data at any node, and if that node does not have this particular object, it will use peer-to-peer network connectivity between nodes to obtain a copy of the data and provide it to the user. It is open, verifiable, and resilient.
Its function allows DID documents to be uploaded onto any node and be referenced by their hash on any node in the network. Modifications to the document modify the hash, so documents are integrity protected by design. Simultaneously, the entire DID storage and distribution mechanism is robust regarding the well-known attacks against Web based identifiers.
In addition, the hash for the document contains no Personally Identifiable Information (PII) and unless we’d make the mistake of adding PII to the DID documents themselves, our design would not expose any kind of PII anywhere.
Of course we were not the first, nor the only ones to realize the potential of IPFS for decentralized identifiers. There has been a prior attempt at using IPFS for DID documents, the IPID DID Method. But it never got much traction, and its use of the InterPlanetary Name System (IPNS) made it less robust. Also, it did not have provisions for the rotation of keys, which is crucial for long-term connections with the same identifier, as well as the ability to switch wallets or upgrade crypto algorithms.
An identifier for Sovereign Data Exchange (SVDX)
The result is did:svdx, our DID method for Sovereign Data Exchange.
Agents generate their active key locally, as well as a key that can be used for the update of the identifier later. The public key of the first key is used as the persistent identifier, creating a persistent Autonomous Identifier (AID).
The second key, which is used for the update of the identifier, is never shared. Only its hash is declared in the document as a next key commitment. Because this key is never actively used until it is time to rotate, it is well protected against being compromised.
Each revision of the decentralized identity documents representing a Decentralized Identifier has a Content Identifier (CID) when stored in IPFS, so the resulting identifier is always the combination of the AID with the CID of the latest revision of the identifier.
Since each revision of the identifier refers back to the previous version by its CID, the result is a sha-256 hash based Key Event Chain of IPFS objects, all the way back to the inception document, the root of the AID in question.
did:svdx:z6MknHKiY477mH97qryHv3zjuHaTLvBbbp6tHS5SvZv67uR4:QmecqVGBxvW7gjffxmYTGFZNPmJcWmYPdD8azB1cZYaY6F
Because the identifier also contains the CID of the current state, starting verification of the Key Event Chain is trivial: Just pull the corresponding object out of IPFS and verify. Check for ancestor, rinse and repeat until you’re at the beginning of the chain. Check whether the AID matches the initial key. Done.
Trivial to implement in web based tool chains
No native IPFS support? No problem. Just pick one of the public IPFS gateways, and with a single request pull the DID document, e.g. https://ipfs.io/ipfs/QmecqVGBxvW7gjffxmYTGFZNPmJcWmYPdD8azB1cZYaY6F.
Thanks to content based addressing, you will get the same document no matter which gateway you use. And you’re welcome to use as many of them as you would like to compare. Although for production use cases it is highly recommended to run your own, which is trivial.
In other words, IPFS allows to integrate classic web based tool chains with decentralized storage and delivery of integrity protected DID documents. It’s as easy as any of the did:web* methods to work with, but does not suffer from the attack surfaces of DNS, TLS and Certificate Authorities.
In addition, it is robust against a number of DDOS scenarios, allows for low impact self-hosting, and eliminates the web server as a central point of attack, surveillance and compromise.
Also, it plays well with DIDComm and other communication protocols, but if you really require web based interaction protocols, they can also be encoded into the identifier. But unlike web based identifiers, exchanging key material via did:svdx mitigates a substantial number of attack scenarios for web connection protocols.
Layering trust
By design did:svdx contains zero personal information. It is deliberately focused on secure key exchange of an Autonomous Identifier, only.
So any relationship starts from a reliable assumption the AID controllers have a strong connection to one another and can maintain it over a longer period of time, including throughout key rotation and changes in cryptography. But they start from zero trust in one another.
Trust is built gradually, through Verifiable Presentations securely exchanged over the connection. Similar to what Christopher Allen describes as “Building Trust in Gradients.”
For SVDX, given it is built for a true P2P, decentralized ecosystem, we surmise that the party initiating a connection first authenticates itself toward the recipient of the connection request before requesting reciprocal information. That should also make data mining or identifier scraping much harder.
Limits of did:svdx
For any design, it is crucial to know its limits. Firstly, the identifier specification does not contain any of the multi-signature capabilities of systems like KERI. Because we did not require it for our use case at hand, we pushed that complexity, along the complexity of secure restore and key rotation, onto the clients — which we control for the use case at hand.
Also, while IPFS plays a role similar to that of Witnesses in KERI, there are no Watchers. So there is no built-in detection of duplicity, as Sam Smith calls it. And while parties can update each other on key rotations using DIDComm, allowing each other to verify they are still talking to the same party, the design has no built-in protections against a controller forking their identity.
For our use case this was not an issue, because there is a central catalogue for the ecosystem to allow looking up the latest, known version of an AID. Which is not ideal for some scenarios. But we considered the solution good enough for what we needed to achieve, given that all controllers need to also maintain their identity and trustworthiness with HIN as the central ecosystem fiduciary.
That said, it should be possible to design a robust duplicity detection on top of did:svdx, and there may even be scenarios where duplicity is not a primary concern as long as agents always ensure to only consider the latest version of an AID authoritative.
So
did:svdxis not a replacement for KERI. But it is a replacement for web based DID methods, offering far better security, and similar efforts of adoption and support. From our own experience we know it took around 6-8 weeks to implement in JavaScript.
What’s next?
The first application using did:svdx in production will have ramped up by April 2025.
By mid 2025 we expect hundreds of thousands of production messages sent each month containing verifiable credentials backed by did:svdx. Our roadmap has us building out additional applications until all the institutions and eventually all the patients in Switzerland will have identifiers within the next 2-3 years.
We have already Open Sourced the initial implementation and will continue to add additional implementations. Also, we would love to finalize the specification so that it can be maximally useful to others. And there may be features that would be required for additional use cases, as well as community-based methods for duplicity detection.
Open questions
- Where is the right place to finalize, publish and maintain did:svdx?
- Who would be interested in participating?
- What are the critical capabilities that may still be missing?
- What kind of best practice operational RFCs should we develop as a community?
If you’re at DICE in Zurich this year, I’d love to sit down and discuss these questions with you — alongside everything else you would like to know about our vision for the Sovereign Data Exchange.
