A single key is a single point of failure in both directions: it can be stolen and it can be lost. Multisignature arrangements remove that by requiring agreement among several keys.
The threshold is the design decision
A multisig is defined by two numbers: how many keys exist and how many must sign. Two of three is common, and larger institutional setups use higher counts.
Raising the required number improves resistance to theft, since an attacker must compromise several independent keys.
Lowering it improves resistance to loss, since more keys can be destroyed without stranding the funds. The two goals pull in opposite directions and the threshold is where they are balanced.
Independence is what makes it work
Three keys held on three devices in the same drawer provide almost none of the intended protection, because one event reaches all of them.
The benefit comes from keys that fail independently: different device types, different locations, different people, different backup methods.
The same reasoning applies to the manufacturers involved, since a flaw in one product line would otherwise affect every key at once.
Implementation differs between chain types
Some chains support multisignature natively in their scripting, so the requirement is enforced by consensus rules and costs little more than a normal spend.
Others implement it in a smart contract that holds the funds and releases them once enough signatures have been collected.
The contract approach is more flexible, allowing spending limits and changeable signer sets, and it introduces contract risk that the native approach does not have.
Recovery information must survive too
Restoring a multisig needs more than the keys. It needs the configuration: how many signers, in what order, and which public keys are involved.
Losing that description can make funds unspendable even with every key intact, which is a failure mode single-key wallets do not have.
Each signer should therefore hold a copy of the full configuration alongside their own key, since it contains nothing secret on its own.
Threshold signatures achieve it differently
An alternative splits one key into shares held by different parties, who cooperate to produce a single ordinary signature without ever assembling the key.
The chain sees a normal transaction, so fees are lower and the arrangement is not visible to observers.
The trade is that the protection is no longer enforced by consensus rules but by the correctness of the signing protocol, which shifts the trust from the chain to the software.