What is an anonymity set?

An anonymity set is the group of people a given action could have come from. The larger it is, the harder it is to tell who did it.

Also called the crowd or traceability set.

Explain it like I’m five

A paper plane lands on the teacher’s desk. If thirty children in the room could have thrown it, the teacher has thirty suspects, and each one has a one in thirty chance of being the thrower. If the teacher learns that twenty-eight of them were out at lunch, only two suspects remain. The suspects are the anonymity set. It shrinks every time the teacher learns a clue, and it is the teacher’s knowledge that decides who is on the list.

What problem it solves

Strong cryptography can hide what a transaction says and still leave clues about who sent it, such as the time, the amount and the address it came from. Anonymity needs a crowd of people who could plausibly have done the same thing, so that the clues do not point to one person. The anonymity set puts a number on that crowd.

How it works, step by step

  • Name the action, such as a withdrawal, and name the observer together with everything the observer knows.
  • List everyone who could have done the action as far as that observer can tell. That list is the anonymity set (Pfitzmann and Hansen).
  • Count it. A set of one means no anonymity, and the largest possible set is the whole system.
  • Check how evenly the suspicion is spread. If the observer finds some members much more likely than others, a large set overstates the protection.
  • Measure the spread with entropy (Serjantov and Danezis). Weight each member by the probability the observer gives them. The entropy, in bits, is how much more information the observer needs to name the person.
  • Watch for clues. Timing, amounts and linked addresses remove members from the list, so keeping them from matching the action keeps the set large.
  • Enlarge the set by making actions look alike, for example by keeping all coins of one kind in a single place, and by waiting, so that more people join before an action.

A worked example

A thousand people could have made a withdrawal, all equally likely. A guess is right one time in a thousand, and the entropy is almost 10 bits. Now take a crowd of 100 where an amount match makes one person 90 percent likely and the other 99 share the remaining 10 percent. The entropy falls to about 1.1 bits, which is what a little over two equally likely people would give. The set has 100 members, and the protection is that of about two.

What decides its strength

  • Size matters, and so does evenness. For a fixed set, protection is greatest when every member is equally likely.
  • The set depends on the observer. Someone who knows more has a smaller set.
  • For an action that has already happened, the set cannot grow later, because people who join afterwards were never candidates. Funds that are still resting do gain from later deposits, since those deposits are in the crowd when the funds move.
  • Clues can only shrink the set, because an observer does not forget what it learns. Zcash’s specification limits its claim about the traceability set to fully shielded transactions, and notes that metadata such as timing or the number of inputs and outputs can still support probabilistic inferences about linkage.

Common misunderstandings

  • More users do not always mean more anonymity. Only users who could plausibly have done the action count, so a different coin, size or time leaves the set.
  • The number of members is not the strength. Uneven suspicion makes the effective size much smaller, as the 100-person crowd shows. Serjantov and Danezis describe a mix whose set includes every sender who ever used it, although older senders carry far less probability than recent ones.
  • An anonymity set belongs to a situation, not to a person alone. It exists relative to what a particular observer knows.

Where it is used

David Chaum introduced the notion in 1988 for dining cryptographers networks, and later work applies it to mix networks. Zcash’s specification calls the equivalent the note traceability set, which includes all earlier notes that the adversary does not control or know to have been spent, and sets it against CoinJoin and CryptoNote, whose sets are smaller because they mix a limited number of transactions. Shielded pools rely on it.

History

Chaum’s 1988 paper used the term. Andreas Pfitzmann and Marit Köhntopp began a shared terminology in 2000, and Serjantov and Danezis showed in 2002 that counting members is not enough and proposed entropy.

In Redacted

Your crowd is everyone who holds the same coin in the Reserve, and with ten of them an outsider who guesses is right one time in ten. The number of different people in that coin matters more than how much they hold. Because the note tree never deletes, every earlier deposit stays a possible source, and new money waits about two hours before it can be used privately. The app shows a privacy label for each coin, worked out in your browser from public chain data.

How privacy works →

Related terms

Sources: A terminology for talking about privacy by data minimization, Pfitzmann and Hansen, version 0.34 · Towards an Information Theoretic Metric for Anonymity, Serjantov and Danezis, 2002 · Zcash Protocol Specification, section 1.2

Updated November 1, 2026. All terms