Sybil resistance is a DAO's defence against one person secretly controlling many identities in order to swing a vote. The name comes from a 2002 paper by John Douceur (itself borrowing the title of a book about multiple-personality disorder), which showed that in any open, permissionless system without a trusted identity authority, a single adversary can forge enough distinct identities to overwhelm honest participants. For a DAO the stakes are direct: the moment voting power is measured by anything other than money – one-person-one-vote, quadratic voting, reputation – a cheap way to mint fake members is a cheap way to seize the treasury.
Plain token-weighted voting is Sybil-resistant by construction: splitting your tokens across a thousand wallets gives you exactly the same total voting power, so fake identities buy nothing. That resistance is bought at a steep price – it is simply plutocracy, where whoever holds the most tokens decides. Every attempt to soften plutocracy by weighting votes more equally – giving each person a say rather than each token – immediately re-opens the Sybil door, because now each additional identity is worth something. Sybil resistance and resistance to plutocracy pull against each other, and bridging them is the central unsolved problem of egalitarian on-chain governance. (a16z governance FAQ)
The dominant answer is proof of personhood: a credential asserting that a wallet belongs to a distinct, unique human, without necessarily revealing which human. The leading approaches trade off differently on privacy, decentralization, and resistance to fakery:
Zero-knowledge techniques increasingly sit on top of these, letting a voter prove "I am a unique registered human" without linking the vote back to the underlying identity. Vocdoni's Ask the World pilot admits each voter on a zero-knowledge passport proof and then takes an encrypted ballot. (Vitalik Buterin: What do I think about biometric proof of personhood?)
Sybil resistance is not an abstract nicety – it is the load-bearing assumption under quadratic voting and funding. Quadratic schemes deliberately make additional influence from a single identity expensive (cost grows with the square of votes), which only dampens whales if each participant is one person. Without a working personhood layer, an attacker splits into many identities and the quadratic penalty evaporates entirely. This is why Gitcoin's quadratic-funding rounds spend heavily on Sybil defence: undetected fake donors don't just steal matching funds, they silently break the mechanism's core fairness promise. (Siddarth et al., "Who Watches the Watchmen?")
The Sybil attacks DAOs actually meet at scale are not attempts to swing a vote – they are attempts to collect a distribution many times over. An airdrop pays per qualifying address, so an attacker who can generate a thousand addresses that each clear the eligibility bar collects a thousand allocations, and no personhood layer stands between them and the money. The defence has therefore been behavioural clustering rather than identity: group addresses by funding source, transaction timing and interaction pattern, and disqualify the clusters.
Two artefacts show how far that got. Hop Protocol opened a public GitHub repository where anyone could file a Sybil report against its own airdrop, paying reporters out of the recovered allocation – crowdsourced detection in place of an identity system. The Arbitrum Foundation then reused the output: its published eligibility rules disqualify outright any address flagged in the Hop bounty, alongside two subtractive heuristics – minus one point if every transaction in the wallet falls inside a single 48-hour window, minus one if the wallet holds under 0.005 ETH and has touched fewer than two contracts.
Both details are worth sitting with. A bug-bounty issue tracker became the industry's shared blocklist, which is what an identity layer looks like when nobody has built one. And the heuristics are deductions rather than exclusions, so a farmer can lose two points and still qualify – a deliberate choice to under-punish rather than strip a genuine user, and an admission that the classifier is not good enough to be trusted with a hard cut.
No personhood system is complete, and each has an honest failure mode. Coercion and vote-selling survive proof of personhood entirely – a verified-unique human can still be paid or pressured to vote a certain way, and a market in rented votes needs no fake identities. Privacy and exclusion cut the other way: biometric and government-ID methods risk surveillance and lock out anyone unwilling or unable to enrol, while looser social methods are easier to game. And every registry faces the base-rate problem that even a small false-acceptance rate, at scale, admits thousands of duplicates. The practical consensus is that Sybil resistance is a spectrum to be raised, not a box to be checked – which is why many DAOs layer personhood scores on top of, rather than instead of, some economic stake.
A caper does not try to verify that each member is a unique human – it makes extra identities pointless instead. Voting weight depends on a member's stake and their demonstrated participation, and participation is tracked by a soulbound token minted by casting ranked ballots and on nothing else, and never transferable between wallets. Anyone may mint their own by voting; nobody can acquire anybody else's. What makes extra identities pointless is how the weight combines the two rather than any barrier to earning the record: splitting one position across several wallets leaves them with less combined weight than the single position had, so splitting is strictly worse than not splitting at all.
The cheap half of a Sybil attack – manufacturing identities – actively costs influence; the expensive half is simply buying more stake, which a caper prices on its bonding curve and does not pretend to prevent. It is a participation gate, not a proof of personhood, and it deliberately trades the egalitarian one-human-one-vote ideal for something a contract can enforce without biometrics or a trusted registry. That same earned weight also sets each member's pro-rata exit claim on the treasury.