Centralised, Decentralised & Federated Identity: A Definitive Architecture Comparison

A deep-dive into the three dominant identity paradigms — centralised, federated, and decentralised — examining their architectures, trade-offs, real-world implementations, and where the industry is heading.

Google Data Center - Photo: Connie Zhou
Feature image: Jhaxhillari / CC BY-SA 4.0 / Wikimedia Commons.

Identity management sits at the core of every modern security architecture, yet choosing between centralised, federated, and decentralised models remains one of the most consequential decisions a security team faces. This article provides a definitive comparison of all three paradigms — examining their underlying mechanics, protocol ecosystems, real-world deployments, and strategic trade-offs that should guide your choice. Whether you are designing a new system or migrating from legacy infrastructure, understanding these identity architectures is essential for building secure, scalable systems.

Which Identity Architecture Is Right for Your Organisation?

The choice depends on your organisation's scale, regulatory requirements, and threat model. Centralised identity offers simplicity and control for small teams. Federated identity enables seamless cross-organisation access through SAML and OIDC. Decentralised identity gives users sovereignty over their credentials using DIDs. Many organisations adopt a hybrid approach, combining centralised IAM with federated external access.

Sources: NIST Identity and Access Management · OAuth 2.0 and OpenID Connect

Identity is the new perimeter. As organisations migrate from on-premises directories to cloud-native ecosystems, the way we structure identity management has become one of the most consequential architectural decisions a security team can make.

There are three dominant paradigms: centralised, federated, and decentralised identity. Each offers distinct trade-offs in control, scalability, privacy, and resilience. Understanding where each excels — and where each breaks down — is essential for anyone designing modern security architectures.

This article provides a definitive comparison of all three models, examining their underlying mechanics, protocol ecosystems, real-world deployments, and the strategic factors that should guide your choice.


1. Centralised Identity

Architecture

In a centralised identity model, a single authority — typically an organisation's own directory service — is the sole source of truth for all identities, credentials, and access policies. Authentication and authorisation decisions are made against this central repository.

The canonical example is Microsoft Active Directory (now Entra ID), but the model also encompasses LDAP directories, RADIUS servers, and monolithic IAM platforms like legacy implementations of PingIdentity or SailPoint before federation was introduced.

The following diagram shows the centralised identity architecture with a single directory serving as the sole source of truth.

flowchart LR
  Users["Users / Devices"] --> Central["Central IdP / Directory (AD, LDAP)"]
  Central --> Resources["Resources (Apps, DBs, etc.)"]

How It Works

  • A user presents credentials (username/password, certificate, or token) to the central directory.
  • The directory verifies the credentials against its stored records.
  • If valid, the directory issues an authentication token (Kerberos ticket, LDAP bind, NTLM challenge-response).
  • Access to resources is determined by policies evaluated at request time against the directory's attribute store.

Key Protocols & Technologies

Protocol Purpose Status
LDAP Directory query and bind authentication Mature, ubiquitous
Kerberos Ticket-based network authentication Enterprise standard (Windows)
NTLM Legacy Windows authentication Deprecated but still present
SAML 2.0 Web SSO assertions Widely deployed in enterprise
OAuth 2.0 / OIDC Modern token-based auth Cloud-native standard

Advantages

  1. Complete Control — The organisation owns every aspect of the identity lifecycle: provisioning, policy enforcement, auditing, and revocation. There is no external dependency.
  2. Simplified Governance — With a single source of truth, policy consistency is straightforward. Access reviews, compliance reporting, and audit trails are contained within one system.
  3. Low Integration Complexity (Internal) — For applications already within the organisation's network perimeter, integration is often straightforward: bind to LDAP, validate Kerberos tickets, or trust an on-premises SAML assertion.
  4. Proven Reliability — Centralised directories have been battle-tested for decades. Active Directory alone runs millions of organisations with predictable behaviour and extensive tooling support.
  5. Strong Accountability — Every identity is directly attributable to an individual or service within the organisation's control, simplifying forensic investigations and regulatory compliance.

Disadvantages

  1. Single Point of Failure — If the central directory goes down, authentication fails across every dependent system. High availability requires expensive multi-site replication and disaster recovery setups.
  2. Limited Cross-Boundary Trust — Centralised directories are designed for internal use. Extending trust to external partners, customers, or acquired companies requires building complex bridge architectures (VPNs, separate tenant directories, identity brokering).
  3. Scalability Constraints — While modern LDAP directories handle millions of entries, the performance characteristics degrade under extreme load. Every authentication request touches the central authority, creating a bottleneck.
  4. Vendor Lock-In — Organisations built on Active Directory become deeply coupled to the Microsoft ecosystem. Migrating away is costly and operationally disruptive.
  5. Privacy Concerns — The central authority has complete visibility into every user's activity across every system. This creates an attractive target for insider threats and external attackers alike.

Real-World Use Cases

  • Large enterprises with on-premises workloads — Banks, government agencies, and healthcare organisations with legacy infrastructure often maintain centralised directories as their primary identity store.
  • Organisations with strict regulatory requirements — Where data sovereignty laws prohibit identity data from crossing borders, a self-contained centralised model is the only viable option.
  • Small-to-medium businesses — Where operational complexity must be minimised and the identity surface area is small, centralised identity is the most pragmatic choice.

2. Federated Identity

Architecture

Federated identity decouples authentication from authorisation by establishing trust relationships between multiple independent identity providers (IdPs). A user authenticates with their home IdP, and that IdP's assertion is trusted by other organisations or services — the service providers (SPs) or relying parties (RPs).

The federated model is the backbone of modern enterprise SSO. When you log into a third-party SaaS tool using your corporate Google or Microsoft account, you are participating in federation.

The following diagram shows the federated identity flow with the trust relationship spanning the user, home IdP, and service provider as a shared domain.

flowchart LR
  subgraph Federation["Federated Trust Domain"]
    direction LR
    User["User"] --> HomeIdP["Home IdP (e.g. Keycloak, Entra ID)"]
    HomeIdP --> SP["Service Provider (SaaS App)"]
  end

How It Works

The most common federation flow (SAML or OIDC) follows this sequence:

  1. User attempts to access a service — The service provider redirects the user to their home IdP's authentication endpoint.
  2. User authenticates with their home IdP — The IdP verifies credentials and issues an assertion (SAML XML) or tokens (OIDC: ID token + access token).
  3. Assertion is returned to the service provider — The SP validates the assertion's signature against the IdP's published metadata (X.509 certificate).
  4. Service grants access — If the assertion is valid and the user has appropriate attributes/claims, access is granted.

Key Protocols & Standards

Protocol Format Primary Use Maturity
SAML 2.0 XML assertions Enterprise web SSO Mature (2005)
OpenID Connect (OIDC) JSON tokens (OAuth 2.0) Modern web/mobile Mature (2014)
OAuth 2.0 Access/refresh tokens API authorisation Ubiquitous
WS-Federation XML tokens Legacy Microsoft ecosystems Deprecated
SCIM 2.0 JSON provisioning Automated user lifecycle Growing adoption

The SAML vs. OIDC Divide

This is one of the most consequential protocol decisions in identity architecture:

  • SAML 2.0 remains the enterprise standard for browser-based SSO. It uses XML assertions signed with X.509 certificates, providing strong guarantees about attribute integrity. However, its XML complexity creates processing overhead and integration friction — particularly for single-page applications (SPAs) and mobile clients.
  • OIDC was designed as a modern replacement. Built on top of OAuth 2.0, it uses JSON Web Tokens (JWTs), which are natively understood by web browsers, SPAs, and mobile apps. OIDC is the default choice for cloud-native applications but lacks SAML's mature attribute assertion model for complex enterprise scenarios.
Rule of thumb: SAML for enterprise web apps and legacy integrations. OIDC for modern web, mobile, and API-first architectures. Many organisations deploy both.

Advantages

  1. Cross-Organisational Trust — Federation enables seamless access across organisational boundaries without sharing credential databases. A partner company can authenticate through its own IdP while the service provider trusts the assertion.
  2. Reduced Credential Management Burden — Users authenticate with a single set of credentials (their home IdP). Service providers never store passwords, eliminating password hash theft as an attack vector.
  3. Centralised Access Control — When a user leaves an organisation, revoking access at the home IdP instantly terminates their ability to access all federated services. No per-service deprovisioning required.
  4. Scalability Through Distribution — Authentication load is distributed across multiple IdPs rather than concentrated in a single directory. Each IdP serves its own user base.
  5. Protocol Maturity — SAML and OIDC are W3C/OSAIC standards with extensive tooling, reference implementations, and commercial support. Keycloak, Entra ID, Okta, and PingIdentity all provide robust federation capabilities.

Disadvantages

  1. Trust Configuration Complexity — Establishing and maintaining trust relationships requires exchanging metadata, managing certificate rotations, and configuring claim mappings. In large federations, this becomes operationally burdensome.
  2. Intermittent Availability — If a user's home IdP is unreachable, they cannot authenticate to any federated service. This creates dependency chains where the reliability of your entire ecosystem depends on external parties' infrastructure.
  3. Privacy Leakage — The home IdP can track which services a user accesses and when (via SAML/OIDC redirect flows). While privacy-enhancing extensions exist, they are not universally deployed.
  4. Attribute Mapping Challenges — Different IdPs use different attribute schemas. Mapping employeeId from one organisation's directory to another's requires custom transformation logic that breaks down at scale.
  5. Security Surface Expansion — Each trust relationship is a potential attack vector. Compromising a single IdP in a federation can cascade across all relying parties. The 2020 Okta breach demonstrated how a compromised identity provider can expose every downstream service.

Real-World Use Cases

  • Enterprise SSO — Employees authenticate once against their corporate IdP (Keycloak, Entra ID) and gain access to dozens of SaaS applications (Salesforce, GitHub, Slack).
  • Acquisition integration — When Company A acquires Company B, federation allows both organisations to maintain separate directories while providing seamless access to combined resources.
  • Education and research — The InCommon and EduGain federations enable thousands of universities and research institutions to share resources through Shibboleth-based SAML federation.
  • Government inter-agency trust — Multiple government agencies maintain separate security domains but establish cross-boundary trust for collaborative operations.

3. Decentralised Identity

Architecture

Decentralised identity (DID) removes the central authority entirely. Instead of relying on a single directory or even a federation of directories, identities are self-sovereign: each entity (person, organisation, device, or AI agent) controls its own identifier and cryptographic keys.

The W3C's Decentralized Identifiers (DID) v1.0 specification (July 2022) and the Verifiable Credentials (VC) data model form the foundational standards. DIDs are URIs that resolve to a DID Document — a machine-readable description of the identity's cryptographic material and service endpoints.

The following diagram shows the decentralised identity flow where the holder presents a verifiable credential to a verifier, with a distributed ledger resolving DID Documents.

flowchart LR
  Holder["Holder (User / Agent)"] --> VC["Verifiable Credential (Issuer signed)"]
  VC --> Verifier["Verifier (Service)"]
  Ledger["Decentralised Ledger / Distributed Key Registry (Blockchain, IPFS, Magma)"] --- Holder
  Ledger --- VC
  Ledger --- Verifier

How It Works

  1. DID Creation — An entity generates a DID (e.g., did:key:z6MkhaXgBZDvotDkL5257faiztiGiC2QtKLGpbnnEGta2doK) and an associated key pair. The DID resolves to a DID Document published on a decentralized network.
  2. Credential Issuance — A trusted issuer (e.g., an employer, government, or educational institution) issues a Verifiable Credential — a cryptographically signed claim about the subject (e.g., "this person is an employee"). The credential contains the issuer's DID, the subject's DID, claims, and a cryptographic signature.
  3. Credential Presentation — The holder (user/agent) presents the credential to a verifier by sharing the signed document. No intermediary is involved — the verifier independently validates the issuer's signature against the issuer's DID Document.
  4. Selective Disclosure — Using advanced VC formats (e.g., BBS+ signatures, ZK-SNARKs), holders can prove specific claims without revealing the full credential — e.g., proving "over 18" without disclosing a birth date.

Key Standards & Technologies

Standard Purpose Status
DID Core 1.0 Decentralised identifier syntax and resolution W3C Recommendation (2022)
VC Data Model 2.0 Structured credential format W3C Recommendation
VC JWT Profile JWT-encoded credentials for web APIs W3C Recommendation
DIDComm v2 Secure messaging between DIDs Decentralized Identity Foundation
ZK-VC (SIS/BBF) Zero-knowledge proofs for selective disclosure Research/early adoption

DID Methods

A "DID method" defines how a specific DID system creates, resolves, updates, and revokes DIDs. Common methods include:

  • did:key: — Self-referential; the key itself encodes the DID. No external registry needed. Ideal for agents and devices.
  • did:web: — Resolves via a standard HTTPS endpoint. Enables easy integration with existing web infrastructure.
  • did:ethr: / did:eth: — Resolves via Ethereum blockchain state. Leverages existing smart contract infrastructure.
  • did:ion: — ION (Identity Overlay Network) by Microsoft — anchors DIDs to the Bitcoin blockchain for tamper-evident resolution.
  • did:peer: — Peer-to-peer DIDs for direct device-to-device trust without any registry.

Advantages

  1. User Sovereignty — Individuals control their own identities and credentials. No corporation or government can unilaterally revoke or modify a user's identity. This is the fundamental philosophical shift from all previous models.
  2. Privacy by Design — Selective disclosure and zero-knowledge proofs enable users to prove specific attributes (e.g., "has a valid driver's license") without revealing the underlying data. Minimal information exposure.
  3. No Single Point of Failure — There is no central directory to compromise. DIDs are resolved through distributed networks, and self-sovereign keys are stored locally by the holder.
  4. Interoperability Across Boundaries — A DID issued by one organisation can be verified by any other party without establishing a federation trust relationship. The verification is cryptographic, not administrative.
  5. Support for Non-Human Identities — DIDs are ideal for AI agents, IoT devices, and service accounts that don't fit the human-centric model of traditional IAM. Each agent can have its own DID, VC portfolio, and autonomous credential lifecycle.
  6. Revocation Resilience — While holders can revoke their own credentials, they cannot be unilaterally revoked by a central authority (unless explicitly designed with revocation registries). This protects against censorship and arbitrary deplatforming.

Disadvantages

  1. Immature Ecosystem — Despite W3C standardisation, production deployments are limited. Tooling, developer documentation, and user experience are still evolving. Most organisations lack the expertise to implement DIDs correctly.
  2. Key Management Complexity — If a user loses their private key, they lose control of their identity. Recovery mechanisms (social recovery, key escrow) introduce centralisation trade-offs that undermine the decentralised model.
  3. Resolution Performance — Resolving DIDs through blockchain or distributed networks is slower than querying a local LDAP directory. For high-throughput authentication scenarios, this latency is a practical barrier.
  4. Legal and Regulatory Uncertainty — Many jurisdictions have no legal framework for recognizing DIDs or verifiable credentials as legally binding identity instruments. The EU's eIDAS 2.0 regulation is beginning to address this, but global alignment is years away.
  5. User Experience Friction — Managing DIDs, keys, and credentials requires a level of technical literacy that most end users do not possess. The "web3 wallet problem" — losing access due to mismanaged keys — remains unsolved at scale.
  6. Limited Attribute Ecosystem — While the standards are mature, the ecosystem of credential issuers is thin. There are few widely-accepted VCs for employment, education, or government credentials compared to the ubiquity of username/password pairs.

Real-World Use Cases

  • Government-issued digital credentials — Estonia's e-Residency program, the EU's Digital Identity Wallet (eIDAS 2.0), and Singapore's Digital Verify initiative are piloting DID-based identity for citizens.
  • Supply chain provenance — GS1's work on Verifiable Credentials for product authentication enables traceable, tamper-evident supply chains.
  • Healthcare credentials — Clinician license verification, HIPAA-compliant patient data sharing, and medical device identity management.
  • AI agent identity — Autonomous agents with DIDs can independently authenticate to MCP servers, API gateways, and other agents without human intermediaries. This is emerging as one of the most compelling near-term use cases.

Comparative Analysis

Decision Matrix

Criterion Centralised Federated Decentralised
Control Organisation owns all Shared between IdP and SP Holder controls identity
Single Point of Failure High Medium (per IdP) Low (distributed)
Cross-Boundary Trust Poor Excellent Excellent (cryptographic)
Privacy Low (central authority sees everything) Medium (home IdP can track) High (selective disclosure)
Scalability Limited by directory performance Distributed across IdPs Limited by resolution network
Implementation Complexity Low Medium-High High
User Experience Familiar (username/password) Familiar (SSO) Evolving (wallets, keys)
Regulatory Compliance Mature framework Mature framework Emerging
Key Management Directory-managed IdP-managed User/holder-managed
Revocation Immediate (directory admin) IdP-dependent Holder-controlled or registry-based
Cost Low-Medium (infrastructure) Medium-High (trust mgmt) High (expertise, tooling)

When to Choose Each Model

Choose Centralised Identity when:

  • You operate within a single organisational boundary with minimal external trust requirements.
  • Your workforce is primarily on-premises or in a private cloud.
  • Regulatory requirements mandate data sovereignty and complete control over identity data.
  • Operational simplicity and proven reliability are higher priorities than cross-boundary interoperability.
  • Example: A regional bank with 5,000 employees using Active Directory for internal systems.

Choose Federated Identity when:

  • You need to share access across organisational boundaries (partners, customers, acquired companies).
  • Your organisation uses multiple SaaS applications and needs unified SSO.
  • You want to reduce credential management overhead while maintaining centralised policy control.
  • Your users expect a familiar login experience (SSO with Google, Microsoft, or corporate Keycloak).
  • Example: A SaaS company enabling enterprise customers to authenticate via their own IdP through SAML/OIDC.

Choose Decentralised Identity when:

  • You need trust across untrusted organisational boundaries without establishing federation relationships.
  • Privacy and user sovereignty are primary design requirements.
  • You are building systems for non-human entities (AI agents, IoT devices) that need autonomous identity.
  • You require cryptographic proof of credentials without relying on a central authority.
  • Example: A consortium of healthcare providers sharing patient consent records via verifiable credentials without exposing the underlying data.

The Hybrid Reality: Most Organisations Use All Three

The binary framing of "centralised vs. decentralised" is misleading. In practice, mature identity architectures layer all three models:

  1. Centralised core — A directory (Active Directory or Keycloak) serves as the primary identity store for employees and devices within the organisation's direct control.
  2. Federated edges — SAML/OIDC federation extends trust to external partners, customers, and cloud services without exposing the central directory.
  3. Decentralised extensions — DIDs and verifiable credentials handle specific use cases where sovereignty, privacy, or non-human identity is required — AI agent authentication, supply chain provenance, or cross-institutional credential exchange.

This layered approach — sometimes called "hybrid identity" — is the pragmatic reality for organisations navigating the transition from legacy IAM to next-generation identity architectures.

The following diagram shows the hybrid identity stack with decentralised, federated, and centralised layers.

flowchart TB
  subgraph Hybrid["Hybrid Identity Stack"]
    Decentralised["Decentralised Layer (DIDs, VCs)"]
    Federated["Federated Layer (SAML/OIDC)"]
    Centralised["Centralised Layer (AD/Keycloak)"]
  end
  Decentralised ~~~ Federated ~~~ Centralised

1. AI Agent Identity as a First-Class Citizen

As autonomous agents become integral to enterprise workflows, the identity landscape must accommodate non-human entities at scale. Microsoft's Entra Agent IDs, CyberArk's Zero Standing Privileges for agents, and SailPoint's unified governance frameworks all point toward a future where DIDs provide the native identity layer for AI systems — enabling agents to authenticate, present credentials, and manage their own access without human intermediaries.

2. Regulatory Push Driving Decentralised Adoption

The EU's eIDAS 2.0 regulation mandates that member states offer digital identity wallets based on verifiable credentials by 2026. This regulatory catalyst is expected to accelerate DID adoption globally, as governments and enterprises align their systems to comply with interoperable credential standards.

3. Protocol Convergence

The line between SAML, OIDC, and VC protocols is blurring. New specifications like the VC JWT Profile enable verifiable credentials to be exchanged using familiar OAuth 2.0/OIDC token flows. This convergence allows organisations to adopt DIDs incrementally — using existing OIDC infrastructure as a bridge to verifiable credential ecosystems.

4. Zero Trust Architecture Integration

Identity is the cornerstone of Zero Trust. The NIST SP 800-207 framework positions identity as the primary trust boundary, and all three identity models are being re-evaluated through this lens. Centralised directories provide strong internal authentication, federation enables zero-trust access to external resources, and DIDs offer cryptographic proof for machine-to-machine trust in microsegmented environments.

5. Decentralised Physical Infrastructure Networks (DePIN)

The intersection of DIDs with physical infrastructure — IoT devices, edge computing nodes, and autonomous vehicles — is creating new identity challenges. Each device needs a verifiable identity that can be independently authenticated by other devices and services without a central authority. DID-based device identity is emerging as the standard solution for DePIN architectures.


Conclusion: Choosing the Right Paradigm

There is no universally "best" identity model. The right choice depends on your organisational context, threat model, regulatory environment, and future trajectory.

  • Centralised identity remains the workhorse for internal enterprise authentication. It is simple, controllable, and proven. But it does not scale beyond organisational boundaries.
  • Federated identity is the bridge between organisations. It enables seamless cross-boundary access while distributing authentication responsibility. Its complexity lies in trust management, not technology.
  • Decentralised identity represents the future of self-sovereign, privacy-preserving digital interaction. While still maturing, its potential to transform how we handle identity — for humans, devices, and AI agents alike — is unprecedented.

The organisations that will thrive are those that understand each model's strengths and deploy them where they add the most value. The future is not a choice between centralised, federated, and decentralised identity — it is the intelligent combination of all three.


Further Reading

  • W3C Decentralized Identifiers (DID) v1.0 — https://www.w3.org/TR/did/
  • W3C Verifiable Credentials Data Model — https://www.w3.org/TR/vc-data-model/
  • NIST SP 800-207: Zero Trust Architecture — https://csrc.nist.gov/publications/detail/sp/800-207/final
  • EU eIDAS 2.0 Regulation — European digital identity framework
  • Rebooting the Web of Trust — https://weboftrust.org/ — Community driving DID adoption
  • SAML vs OIDC: OneLogin's Comparison Guide — https://www.onelogin.com/learn/oidc-vs-saml/

Feature image: Jhaxhillari / CC BY-SA 4.0 / Wikimedia Commons.