← all systems · see it in The System
Digital identity
A digital identity is data stored on computer systems relating to an individual, organization, application, or device (src:wp:digital-identity). NIST: 'the unique representation of a subject engaged in an online transaction', always unique in the context of a digital service but not necessarily across contexts, and 'the online persona of a subject', whose single definition 'is widely debated internationally' (src:nist-sp800-63-3). Walked as the whole practice of representing and recognising people, organisations, devices and software online: attributes and footprints, identifiers, credentials, identity proofing and assertion, the providers, federations, registries and ledgers that vouch, the persona a service, a state or a crowd takes the subject to be, and its failures: re-identification, Sybil identities and identity theft.
Properties published scienceour proposal
Sourced fields, and our own readings.
| born at | the 1960s, with time-shared computers and the first password logins (authentication walk, CTSS 1961); the term identity theft coined 1964 |
| born precision | decade |
| boundary | Inside: the subject's digital representation and everything that makes and keeps it: attributes, preferences, traits and the footprints of activity; identifiers unique within a scope (usernames, email addresses, URIs, UUIDs, DIDs, subject identifiers); credentials that bind an identifier to authenticators; identity proofing (who someone is), authentication (control of the authenticators) and federation (assertions conveyed to relying parties), as NIST separates them into IAL, AAL and FAL; the providers, registries, federations, wallets and ledgers that issue, hold and vouch for identities; the profile or data double assembled from scattered footprints; and the persona the subject is taken to be. Outside, as neighbours: authentication as its own system (walked; this walk points at its layers), access control (what an identified subject may do), the cryptography it borrows (digital signatures, public-key cryptography, hash functions), privacy as a value and a field (not walked), the people and machines represented, the thieves and trackers who use the representation without consent, and the law that gives identities legal effect. |
| nested level | A technology (rung 12) carried by people, firms and states (rung 11) on computers and the internet. Part of information security and identity and access management; leans on authentication, digital signatures and public-key cryptography; feeds access control, commerce, government services and social life online. |
| scale | technological |
Measures (16) published science
| characteristic clock: browser fingerprint drift | 45.52% of fingerprints changed within one day; a device linked for 51.8 days on average; 26% tracked over 100 days |
| per-hand-off efficiency: fingerprint carried from one day to the next unchanged | 54.48% (100% - 45.52%) fraction unchanged after one day |
| unique browser fingerprints, Panopticlick 2010 | 83.6% (94.2% with Flash or Java), of 470,161 fraction |
| unique browser fingerprints, 2018 (about 2 million from one French site) | 33.6% (18.5% on mobile) fraction |
| mobility unicity: points needed to single out 95% of 1.5 million people | 4 spatio-temporal points (2 points: over 50%) |
| uniqueness against resolution | decays approximately as the 1/10 power exponent |
| re-identification with 15 demographic attributes (US) | 99.98% fraction correctly re-identified |
| likely unique on 5-digit ZIP, sex and birth date (US, 1990 census) | 87% (216 of 248 million) fraction |
| identity theft prevalence, US adults | about 10% in 2016; 7% in 2012 fraction per year |
| victims with severe emotional distress | 1 in 10 (about 2.6 million) fraction |
| FTC identity theft reports, 2024 | over 1.1 million; 18% of all Consumer Sentinel reports reports per year |
| breach-to-theft yield in one card breach | about 1,800 instances from four million records (about 0.045%) fraction |
| people without a valid identity credential | about 850 million people |
| people for whom official ID is problematic | about 1.1 billion (2017 estimate cited) people |
| UUID size | 128 bits |
| behavioural identity accuracy (encyclopedia claim) | up to 95% accuracy |
Open contentions (4) open contention
Where sources or readings disagree. Both stay on the map.
- C-1636 Can data be made anonymous enough to share safely?
booked: The map books how often people were picked out again in the studies (87%, 95%, 99.98%) and keeps the release-and-forget practice as what is actually done.
the other reading: de-identified and sampled data are anonymous enough to release (the release-and-forget practice the papers describe); even heavily sampled de-identified data are unlikely to meet GDPR's anonymisation standard (Rocher 2019) - C-1643 Do data breaches cause identity theft?
booked: The map keeps both: surveys link breaches to every kind of identity theft, and most breaches are not followed by detected theft.
the other reading: data breaches drive identity theft (Burnes et al. 2020: associated with every subtype); most breaches do not lead to detected identity theft (GAO via the encyclopedia) - C-1649 What is a digital identity?
booked: The map keeps three readings on three floors: the data about you, the identifier used in a transaction, and the persona you present.
the other reading: digital identity is the data about a subject; digital identity is the unique representation used in a transaction; digital identity is the persona: a social identity shaped by group memberships; digital identity is a viewpoint within a mutually agreed relationship rather than an objective property (encyclopedia, uncited) - C-1683 Does self-sovereign identity put people in control of their identity?
booked: The map books its promise, its critique, and the proof that without some trusted authority one machine can pose as many.
the other reading: SSI removes reliance on central parties ('classic Identity Management registration processes can be replaced'); EU-wide SSI shifts existing power balances rather than removing them, and puts individuals in a vulnerable position (Giannopoulou 2023); without a logically centralised authority Sybil identities are always possible (Douceur 2002); proof-of-personhood protocols claim a subjective way round it (Siddarth 2020)
Partners and parts published science
Built from / builds (7)
- Authentication
- Identity proofing
- Attributes, preferences, traits and footprints
- Digital identifiers
- Credentials
- Federation and assertions
- Identity providers and credential service providers
Sits inside / holds (3)
Trades with (valence) (17)
- Phishing via sys:internet · Digital identity loss
- Privacy via sys:human-society · Digital identity mixed
- Social engineering via sys:internet · Digital identity loss
- Surveillance via sys:human-society · Digital identity loss
- Human (Homo sapiens) via sys:human-society · Digital identity gain
- State (polity); the nation state its predominant modern form via sys:human-society · Digital identity gain
- Law (rules and law) via sys:human-society · Digital identity gain
- Internet via sys:human-society · Digital identity gain
- Human society (from local community up to societies) via sys:human-society · Digital identity gain
- Computer via sys:human-society · Digital identity gain
- Cryptographic hash function via sys:human-society · Digital identity gain
- Public-key cryptography via sys:human-society · Digital identity gain
- Digital signatures and certificates via sys:human-society · Digital identity gain
- Authentication via sys:human-society · Digital identity gain
- Access control via sys:human-society · Digital identity gain
- Threat actor via sys:human-society · Digital identity loss
- Organization via sys:human-society · Digital identity gain
expressed by (17)
- Generation/Synthesis
- Recombination/Composition
- Templating / Standardization
- Interoperability/Transduction (Bridging)
- Encapsulation & Modularity (Abstraction)
- Distributed Autonomy/Decentralized Execution (Subsidiarity)
- Sensing/Scanning
- Multi-Scale Data Fusion / Triangulation
- Trail-Marking / Externalized Memory (Stigmergy)
- Surprise / Change-Point Detection
- Reactive Response
- Physical/Virtual Barrier
- Technique-Based Identification (a.k.a. Technique-Based Detection)
- Validation of Identity/Intention
- Enclave/Segregation
- Resilience & Repair
- Deterrence
floor filled by (18)
- Surveillance
- Human (Homo sapiens)
- State (polity); the nation state its predominant modern form
- Law (rules and law)
- Internet
- Human society (from local community up to societies)
- Computer
- Public-key cryptography
- Digital signatures and certificates
- Authentication
- Organization
- Identity proofing
- Attributes, preferences, traits and footprints
- Digital identifiers
- Credentials
- Federation and assertions
- Identity providers and credential service providers
- Self-sovereign identity (DIDs and verifiable credentials)
represents (1)
Kinds (6)
- Federation and assertions
- Self-sovereign identity (DIDs and verifiable credentials)
- Isolated or centralised accounts (one identity per service provider)
- National digital identity systems (Aadhaar, eIDAS wallets, national smart cards)
- Machine and device identities (UUIDs, certificates, device fingerprints)
- Online identity (the social persona)
Notes published science
- Four places and times are enough to single out 95% of 1.5 million phone users from their location records, and blurring the data barely helps: uniqueness falls only as the tenth root of the resolution. Fingerprints, for comparison, were taken in 1930 to need 12 points (src:demontjoye-2013-unique-crowd).
- The same browser fingerprint does two opposite jobs: banks and shops use it to spot fraudsters and bots, trackers use it to follow you; and it drifts, with 45.52% of fingerprints changing within a day, yet one study's algorithm still followed devices for 51.8 days on average (src:laperdrix-2020-browser-fingerprinting).
- In 2002 Douceur proved that, without some trusted authority vouching that one identity is one entity, a network can never stop one machine from pretending to be many (the Sybil attack), except under unrealistic assumptions; the self-sovereign identity movement and 'proof of personhood' protocols are still working around that proof (src:douceur-2002-sybil; src:siddarth-2020-proof-of-personhood).
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
digital identity as practitioners describe it: subjects in the world up to the persona a service, a state or a crowd takes them to be (reader's view)
- L1: Subjects in the world: people, organisations, devices, networks, software and services, unseen at the other end of the wire
- L2: Data about the subject: attributes, preferences, traits, the footprints of activity
- L3: Identifiers unique within a scope: usernames, email addresses, URIs, UUIDs, DIDs, subject identifiers; public or pairwise, resolvable or not
- L4: Credentials and certificates binding identifiers and attributes to authenticators
- L5: Assurance and profile: identity proofed (IAL), authenticated (AAL), asserted (FAL), scored; the data double
- L6: Trust networks: identity providers, federations, certification authorities, national systems and decentralised ledgers
- L7: The persona: who the subject is taken to be, by a service, a state or a crowd; legal and social identity
the boardroom (shared floors)
- L5: Assurance: how sure the service is about who or what is asking (proofing, authentication and federation strengths, risk scores, Sybil checks), and how much it needs to know (pseudonymous where possible)
- L5: The record and the profile: database identity, transaction identity and the data double assembled from footprints, fingerprints, linked datasets and self-presentation
- L6: Trust networks: identity providers, federations, certification authorities, national systems and ledgers that vouch for subjects to relying parties (C vouches to A for B)
- L7: The persona: who the subject is taken to be online, by services (one identity per context), by states (a legal identity) and by crowds (a social identity)
way in: Assertions from an identity provider (federation, single sign-on, OpenID Connect, SAML)
- L1: An authentication event at the identity provider
- L2: Packed into an assertion signed by the IdP (FAL1)
- L3: Carried to the relying party; one ticket trusted across many systems; the subject identifier local to its issuer
- L4: The relying party checks signature, issuer and subject; at FAL3 also proof of a key
way in: Authenticators presented at each return visit (the authentication walk's channels)
- L1: An authenticator the subscriber holds (a secret, a token, a key, a body feature)
- L2: Proved in an authentication protocol
- L3: Return visits tied to the same subscriber
- L4: Checked against the credential that binds the authenticator to the identity
way in: Behaviour and footprints (logins, purchases, locations, histories)
- L1: The subject's activity: searches, purchases, logins, places
- L2: Logged as events with device and environment variables
- L3: Entity relationships of a new event compared with past events
- L4: Scored against tolerances into a transaction risk score
way in: Device and browser fingerprints
- L1: The device's configuration: headers, fonts, screen, plugins, graphics
- L2: Read by scripts, storing nothing on the device; defenders add noise
- L3: Combined into a fingerprint that drifts (45.52% changed within a day)
- L4: Matched against fingerprints seen before (83.6% unique in 2010, 33.6% in 2018)
way in: Identity evidence (passports, licences, civil records) for identity proofing
- L1: Documents and registers issued by states and other issuing sources; about 850 million people have none valid
- L2: Presented remotely or in person; IAL3 needs physical presence
- L3: Validated with the issuing source
- L4: Proofing decision: the applicant is who they claim to be; enrolled as a subscriber
way in: Law, regulation and guidelines (GDPR, eIDAS, anti-money-laundering, NIST)
- L1: The state's sovereign responsibility for identity systems
- L2: Statutes and rules: anti-money-laundering reliance, GDPR, eIDAS
- L3: Guidelines that set requirements per area
- L4: Rights exercised: data deleted on request
way in: Machine and service identities (UUIDs, device keys, certificates)
- L1: Devices and services that act as subjects, often in several roles
- L2: An identifier minted without a registry (UUID) or a key pair generated
- L3: A certificate binding the key to a name
- L4: Checked when a server signs in with its certificate
way in: Other datasets and outside information (quasi-identifiers, mobility traces)
- L1: Datasets released after de-identification and sampling
- L2: Quasi-identifiers: ZIP code, sex, date of birth; places and times
- L3: Joined with outside information; hashing identifiers is the common mitigation
- L4: Unicity: a few points single a person out (4 points, 95%; 15 attributes, 99.98%)
way in: Other people vouching (proof of personhood, social graphs)
- L1: Other members who know the subject
- L2: Vouches that one living person stands behind one identity, against many virtual personas
- L3: Recorded on a registry, disputes taken to a court
- L4: Checked for uniqueness and singularity: one human, one identity
way in: Self-presentation and group membership (profiles, posts, avatars)
- L1: The person's offline and online group memberships and sense of self
- L2: Shown as a profile, posts, a username or an avatar, from verified match to role-play
- L3: Shaped by the platform: anonymity, pseudonymity and verification
- L4: Read by audiences at once (context collapse)
way in: Signs of misuse (accounts and charges a subject never made; breach notices; reports)
- L1: Another person uses the identity: accounts, charges, tax returns, care in the victim's name
- L2: Noticed on statements and credit reports, often late; victims often do not know how their data was obtained
- L3: Reported to agencies (FTC Sentinel: over 1.1 million identity theft reports in 2024); hours spent resolving
- L4: A theft report and plan: accounts reviewed, multi-factor authentication used
way in: Verifiable credentials and decentralised identifiers presented from a wallet
- L1: An issuer attests to attributes of the subject
- L2: Held by the subject in a personal wallet
- L3: Presented with a DID that resolves to its document
- L4: Verified against keys in the DID document
way in: What the subject says about itself (names, usernames, attributes, passwords)
- L1: The subject's own name, choices and secrets, held in memory (or a fabricated set: synthetic identities)
- L2: Typed into a registration or profile form; kept up by the subject (changing passwords is protective)
- L3: Stored as attributes bound to an account identifier
- L4: Accepted as self-asserted at IAL1: no link to a real-life identity required
way out: Access and transactions for the identified subject
- L1: Limited access granted on the subject's behalf
- L3: Handed to authorization: what this identity may do
- L5: Assurance enough for the risk: access allowed
way out: Assertions sent to relying parties
- L1: Carried to the relying party
- L3: Only what is needed: a property asserted, a pairwise identifier per relying party
- L6: The identity provider issues an assertion
way out: Duplicate and false identities rejected
- L1: Fewer false personas on the network
- L4: The duplicate removed; its vouchers lose their deposit
- L5: A duplicate found
way out: Identities and credentials issued
- L1: A DID and its document published
- L4: The credential binds identifier, attributes and authenticator
- L6: A trusted party issues the credential or certificate
way out: Repair after misuse: accounts closed, credentials revoked, data deleted
- L1: Data deleted on request; accounts closed after death
- L4: Recovery work; distress for one victim in ten
- L5: Misuse recognised
way out: Tracking and re-identification
- L1: A named person found in a released dataset
- L4: A device followed across fingerprint changes (51.8 days on average)
- L5: The profile linked across services
way out: What the subject is shown (personalised content and offers)
- L1: Shown on the subject's screen
- L4: Offers and content tailored
- L5: The profile read