Why DNA-based proof of personhood doesn't work — and what helixstreet does instead

A recurring pitch in the DeSci-meets-identity corner of the industry says: your genome is unique, so it makes a great Sybil-resistance primitive — use DNA as proof of personhood. It doesn't, and we want to be on the record about why.

This note is a longer companion to the position we take on the homepage. We owe readers a clear, citation-backed walk-through, because the alternative is that someone reads a half-coherent pitch deck and shows up expecting helixstreet to power a "DNA-PoP" product we will never build.

The bar a proof-of-personhood primitive has to clear

A workable PoP scheme has to satisfy a tight set of properties at the same time. Boiled down (see Buterin's 2023 essay on biometric PoP [1] and the 2025 Personhood Credentials taxonomy paper [2]):

• Unique-per-human — no two distinct humans should map to the same identifier; no human should be able to mint two.
• Live-capture — the credential must be derivable only with the person physically present at issuance.
• Revocable — on compromise, the user must be able to issue a new credential and invalidate the old.
• Non-coercive — capture should not leak information that could be used against the holder or family later.
• Sybil-resistant under bulk collection — an attacker who scrapes a lot of source material at scale should not be able to mint identities.

Worldcoin/World [3] chose iris-recognition because iris satisfies (roughly) all five: ~250 bits of entropy, live capture in front of an Orb, the iris template is hashed and the hash is the only thing that leaves, and the iris pattern is fixed but private (you can't scrape someone's iris off a coffee cup). One can argue about the centralization of the issuance, the orb hardware, the geopolitics — but the underlying primitive isn't broken.

How DNA fails each requirement

"DNA is unique" is true. That is the only PoP property it satisfies. Every other property — live-capture, revocability, non-coercion, Sybil resistance under bulk scraping — it fails. And the failures aren't fixable with more cryptography: they're properties of the substrate.

Non-revocable biometrics are the strongest possible identifier and the worst possible credential. DNA is the least revocable biometric we have.

Why the cryptographic dodge doesn't save it

A natural response is: "fine, don't store the DNA — store a zero-knowledge commitment over it." Two problems.

First, the commitment is still derived from the genotype. Anyone who later obtains the user's DNA (a relative submitting to 23andMe, a discarded coffee cup, a court order) can recompute the commitment and impersonate the holder. Iris hashes don't have this property because iris isn't shed.

Second, the family-tree leak is structural, not cryptographic. If Alice mints a DNA-based PoP credential, then Alice's brother's later genotype submission tells the world (probabilistically) that Alice's credential exists and roughly who issued it. ZK over a commitment does not fix this; the leakage happens at the population-genetics layer.

The cleanest reference here is the academic literature on re-identification from partial SNP sets — the Gymrek [4] and Erlich [5] results are still the relevant counterexamples in 2026.

What helixstreet does instead

The properties people want from "DNA-PoP" usually decompose into two different things: (a) proof that a user has a genome with certain properties (eligibility for a trial, ancestry for a Clan, kinship for a family-finder service), and (b) proof that a user is a unique human. These are different problems and they take different tools.

(a) Eligibility and ancestry — we can do this

For the first set of properties, Icefish (IACR 2026/463) is the current state of the art [6]: practical zk-SNARKs for narrow genomic operations (sequence alignment, GWAS verification, CRISPR target match). Sub-20-minute proving for literature-scale GWAS, 30× faster sequence-alignment proofs than prior art. This is the layer helixstreet builds against. The chain stores the verifier; the user generates the proof off-chain against their local VCF; the chain records this proof was verified at this block, not the proof inputs.

For kinship, the credible work is HE-based, not zk-based: SF-Relate (Genome Research, 2024) [7] runs federated relative-discovery via multiparty homomorphic encryption at UK-Biobank and All-of-Us scale. The chain's role there is coordination — sessions, keys, consent receipts — not computation.

(b) Proof of unique humanity — we don't do this, and we recommend you don't use DNA for it

For Sybil resistance specifically, helixstreet defers to credentials issued by external personhood systems — iris-based World ID, government-issued personhood credentials per the 2025 PCred taxonomy [2], or human-review issuance via KILT registrars. These are credentials that do meet the bar; the chain consumes them over XCM rather than minting its own broken alternative.

What the honest pitch looks like

helixstreet is a coordination and audit layer for consented genomic data and the workflows around it. It does not put genomes on chain, it does not issue identities from genomes, and it does not claim that any zk construction makes those things safe. It does verify narrow eligibility proofs, anchor consent receipts, and coordinate federated kinship sessions, and it does compose with the personhood systems that already work.

If a pitch you've seen claims "DNA proof of personhood on Polkadot" or anything adjacent, it isn't us, and we'd quietly suggest it isn't anyone else either.