Every platform issues its own identity and none of them talk to each other. The result is a specific and familiar set of costs: you upload the same documents for every service, each service stores its own copy of your sensitive data, and none of the verification transfers. The platforms accumulate honeypots. You accumulate friction.
Decentralised identity is the attempt to fix that by separating the claim from the document, and by putting the user rather than the platform at the root of trust.
Three primitives
Decentralised identifiers give a subject a persistent handle that resolves to control of a key rather than to a row in someone's database. Nothing about the identifier says who you are; it says who can prove control of it.
Verifiable credentials are signed statements. An issuer signs a claim — over 18, resident in a jurisdiction, accredited investor, employed by X — and any verifier can check the signature without contacting the issuer. The W3C has standardised the data model, which matters more than any single implementation, because it is what makes credentials portable between ecosystems.
Attestations are the on-chain form of the same idea. Ethereum Attestation Service makes a claim cheap to register and cheap to read, turning "this address was verified by that issuer" into a queryable primitive. It is the most widely used of the three precisely because it is the least ambitious.
The three roles to keep straight
Reliable identity systems separate three parties, and conflating them is where most designs fail.
An issuer vouches for a claim. A holder stores it. A verifier checks it. Between holder and verifier sit zero-knowledge proofs, so the verifier learns the predicate and nothing else — that you are over 18, not your birth date.
That architecture satisfies a compliance officer and a privacy engineer at the same time, which is unusual, and it is why it is the one being standardised. It also makes the liability question sharp: a false attestation has two candidate owners, the issuer who signed it and the verifier who accepted it, and most legal frameworks have not decided between them.
Soulbound tokens, and why they are harder than they look
A soulbound token cannot be transferred. The appeal is obvious: attendance records, contributor badges, and governance participation all work better when they cannot be bought, and a non-transferable badge makes sybil resistance cheaper because farming it requires actually doing the thing.
Three problems are structural rather than implementation details.
Permanence is the wrong default. An on-chain badge recording your entire history is also a permanent, public, linkable record of that history, which becomes a privacy problem the moment two badges share an address.
Revocation is missing. A credential issued by mistake, or by an issuer later found to be compromised, cannot be clawed back from a self-custodied wallet. ERC-5484's consensual soulbound tokens address part of this by encoding a burn right into the token itself, but the general case is unsolved.
Linkability survives. Because the token lives at an address, "non-transferable" does not prevent the wallet from being sold — it only makes the sale harder to enforce.
The practical rule that follows: keep personal credentials off the public ledger, and put a proof of them on-chain only when a verifier actually needs to check one.
Sybil resistance is the hard part
Everything above proves a claim. None of it proves that a claim belongs to a distinct human, and most governance attacks are not about false credentials — they are about one person holding a thousand valid ones.
Worldcoin attacks this with iris scanning and a uniqueness proof. Polygon ID issues ZK credentials tied to verification. Gitcoin Passport aggregates many weak signals into a score. Each is a different bet: Worldcoin on biometric uniqueness at the cost of a hardware rollout and serious privacy objections; Polygon ID on issuer trust; Passport on heuristics that an attacker can farm cheaply if the signals are public.
There is no clean answer. The honest framing is that sybil resistance is a cost imposed on attackers, not a property that can be proven, and the design question is always "how much does it cost to manufacture 1,000 identities" rather than "is this person real".
The regulatory tension nobody has resolved
Two conflicts are unresolved by technology.
GDPR versus immutability. A right to erasure cannot be satisfied by a ledger that does not forget. This is why personal data has to stay off-chain and only proofs go on it — a design constraint that follows from the regulation, not from cryptography.
Linkability at the verifier level. A user who presents the same credential to five verifiers has reconstructed the fragmentation problem in a new form, because the five can now correlate. Persistent identifiers make this worse and are convenient for exactly that reason.
The implementation bugs to watch are equally mundane: replay (a proof reused outside the context it was issued for) and issuer collusion (two issuers correlating their attestations about one subject). Both are solved by binding proofs to a verifier and a nonce, and both are routinely skipped.
References
Next, read Privacy Protocols 101. More in Security & Privacy.