A multisig converts one catastrophic failure into several ordinary ones. Lose a key in a single-key wallet and everything is gone; lose a key in a 3-of-5 and you still hold the funds. That is the whole safety argument, and it is why the failure modes of multisigs are operational rather than cryptographic: the contract usually works, and the people around it usually do not.
Choose the threshold for the accident you expect
An M-of-N configuration trades safety against the ability to act. Two of three is the smallest setup worth calling multisig: routine, workable, and fails if one signer vanishes and another is compromised. Three of five is the common choice for treasuries — two losses or two absences tolerated, still quorumable within a day. Five of nine and above belongs to organisations with paid operational staff, because coordination itself becomes the risk.
The number to avoid is a threshold of one relative to quorum-1 in the same place. A 2-of-3 where two keys sit on the same person's laptop is a single-key wallet with extra steps.
Pick signers who cannot fail together
Signer selection is where most of the security is decided, and it is not a technical question. Do not choose people who share an employer, a jurisdiction, a home, an exchange account, or a threat model. A signer set that can all be reached by the same legal order, or that all depend on the same cloud provider, is one signer duplicated.
Practical diversity looks like: different countries, at least two independent hardware wallets from different vendors, at least one signer outside the founding team, and documented replacement plans for each.
Never sign a transaction you cannot explain
Blind signing is the operational failure that drains multisigs. Before approving, each signer should be able to say what the transaction does, who receives what, whether the target contract is verified, and whether the value and the destination match what was discussed. If a signer cannot, the correct action is to pause.
That rule needs infrastructure to be enforceable: propose transactions with human-readable intent, use simulation where the wallet supports it, and put a delay between proposal and execution so a mistake has time to be noticed. On Safe, transaction guards and role-based permissions exist for exactly this reason.
Treat recovery as a practised procedure
The scenario to plan for is not theft — it is that one signer loses their device, dies, or becomes unreachable during a volatile week. Write down who replaces them, how the swap is executed, and what quorum is required to change the set. Then test it: recover one signer's key from backup in a rehearsal and confirm the address. A recovery plan that has never been executed is a document, not a plan.
What DAOs get wrong
A multisig holding a treasury is not governance, it is custody. The healthy pattern separates the two: a council that executes decisions already made through a proposal process, with published signers, published thresholds, and a rule for emergencies that includes an obligation to disclose afterwards.
Transparency is the cheapest control you have. An anonymous multisig holding community funds asks for trust it has not earned, and it makes every future proposal a negotiation about whether the council can be believed.
Review on a schedule
Run a quarterly check: are the signers still who you think, is the threshold still right for the amount held, are the keys still on hardware, are the backups where they should be, and has anyone left the project without being removed from the set? Most losses in multisigs are discovered during a review that happened late.
Next: the on-chain side — how slashing risk changes staking and restaking decisions. Join our Discord.
References
Next, read Hardware Wallets Security 101. More in Security & Privacy.