Capability Certificate

A signed certificate binding a public key (the subject) to an authority (a description of what actions the subject is permitted) — the SPKI-style realisation of Capability Security in a distributed cryptographic setting. Composes by chain (concatenated delegations, with authority intersected at each step) and by attenuation (a delegate can issue a sub-certificate granting strictly fewer rights). Modern descendants: Macaroons (Birgisson et al. 2014), zCaps (W3C Authorization Capabilities), object-capability tokens in cap-secure RPC.

In this vault

Backlinks

Linked Pages

Macaroons

Birgisson, Politz, Erlingsson, Taly, Vrable & Lentczner’s (2014) lightweight authorisation tokens — HMAC-chained capability tokens supporting attenuation (the holder can derive strictly less-permissive sub-tokens) and third-party caveats (delegating verification of a condition to a separate service). The most-deployed modern descendant of SPKI/SDSI capability certificates, used in production at Google and elsewhere. A practical realisation of Capability Security in HTTP-API authorisation.

In this vault

Capability Security

Access-control style where authority is conveyed by unforgeable object references; lexical scope in languages like Scheme suffices as a capability kernel.

In this vault

SPKI-SDSI

SPKI / SDSI: A Simple Distributed Security Infrastructure with Linked Local Names

Reference: Rivest, R. L. & Lampson, B. (1996). SDSI — A Simple Distributed Security Infrastructure. (Working document, MIT.) Followed by: Ellison, C., Frantz, B., Lampson, B., Rivest, R., Thomas, B. & Ylonen, T. (1999). SPKI Certificate Theory. RFC 2693, IETF. (Also: Clarke, D. E., Elien, J.-E., Ellison, C., Fredette, M., Morcos, A. & Rivest, R. L. (2001). Certificate Chain Discovery in SPKI/SDSI. Journal of Computer Security 9(4).) Open access (RFC 2693) · SDSI 1.0 (Rivest-Lampson, MIT) · Chain discovery (Elien MIT MS thesis 1998)

Summary

Rivest, Lampson, Ellison and the SPKI working group develop SPKI/SDSI, a public-key infrastructure that takes a sharply different stance from the X.509 / hierarchical-CA tradition: instead of a global namespace of certified identities rooted in a chain of certificate authorities, SPKI/SDSI gives each principal its own local namespace and certificates that bind keys to authorisations rather than identities. SDSI’s central innovation is linked local names: K_alice's friends is a name in K_alice’s namespace; K_alice's friends's secretary is a chain of links — no global registry needed. Authority delegation is similarly local: a certificate K_a → K_b : auth issued by K_a grants authority auth to K_b, optionally with delegation rights. Authority is what matters, not who — a distinction Lampson summarised as “we’re moving from authentication to authorization.” The SPKI Certificate Theory (RFC 2693) formalises this: certificates are 5-tuples <Issuer, Subject, Delegation, Authority, Validity>; chain discovery (Clarke et al. 2001) gives a sound and complete algorithm for deriving “does this principal have this authority?” from a collection of certificates. The design directly anticipates modern decentralised identity / object-capability composition: SPKI/SDSI is the conceptual bridge between the object-capability tradition (E, Spritely) and the cryptographic-identity tradition (X.509, PGP). DIDs and Verifiable Credentials inherit substantial conceptual machinery from SPKI/SDSI; Macaroons (Birgisson et al. 2014) are SPKI-style attenuable capability tokens for HTTP APIs.

Key Ideas

Authorization, not authentication: certificates bind keys to authorities — what the principal is permitted to do — rather than to identities. The question “who is this key?” is replaced by “what is this key allowed to do?”
Local namespaces (SDSI): each principal has its own namespace; names are local identifiers like K_alice's bob, with no global resolution. Linked local names chain across namespaces: K_alice's bob's friends resolves through K_alice’s and then K_bob’s namespaces.
Delegation as a first-class certificate field: a certificate may grant authority with or without the right to re-delegate; chains of delegated certificates are the principal authority-derivation mechanism.
5-tuple certificates: <Issuer, Subject, Delegation-flag, Authorization-S-expression, Validity> (RFC 2693 §2.4 calls the field “Authorization”, not “Authority”); the Authorization is an S-expression in a structured language (action, object, qualifier) that supports rich access-control specifications.
Chain discovery (Clarke et al. 2001): given a goal K_principal has authority A, the chain-discovery algorithm searches the certificate collection for a derivation; sound and complete; polynomial in the size of the certificate collection for many practical fragments.
No mandatory identity binding: certificates may bind keys to local names (SDSI-style) or directly to authorisations (SPKI-style) — identity is just one kind of authority. The system can be used without ever assigning a global “real name” to a key.
Capability composition: SPKI authorities compose under intersection and chain; the resulting authority is the most-restrictive intersection along the chain. This is exactly the capability-security composition rule lifted to a certificate-based distributed setting.

Connections

Conceptual Contribution

Claim: A distributed security infrastructure should bind cryptographic keys to authorisations (what they may do) rather than to identities (who they “really” are); each principal should have its own local namespace; authority delegation should be first-class with an explicit chain-discovery algorithm. The X.509 hierarchical-CA model is the wrong fit for most distributed-system access-control needs.
Mechanism: 5-tuple certificate format (issuer / subject / delegation / authority / validity); local namespaces with linked-local-names resolution; sound and complete chain-discovery algorithm; authority composition under intersection.
Concepts introduced/used: SPKI, SDSI, Linked Local Names, Capability Certificate, Authority-not-Identity, Chain Discovery.
Stance: foundational systems-design paper / IETF specification.
Relates to: Direct conceptual bridge between the object-capability tradition (Dennis & Van Horn 1966, EROS, the E-language line in Miller 2006) and the cryptographic-identity tradition (X.509, PGP). SPKI/SDSI explicitly takes the capability stance — unforgeable name = capability — and lifts it into a distributed cryptographic setting where the unforgeability comes from the public-key cryptography rather than from memory protection. The Macaroons paper (Birgisson, Politz, Erlingsson, Taly, Vrable & Lentczner 2014) is SPKI/SDSI’s most-deployed modern descendant: HMAC-chained authority tokens with attenuation, used in production at Google and elsewhere. Modern decentralised-identity standards — W3C DIDs, Verifiable Credentials, the Authorization Capabilities (zCaps) work — inherit substantial conceptual machinery from SPKI/SDSI even when they do not cite it directly. Within this vault, SPKI/SDSI is the missing distributed-cryptographic counterpart of the in-process capability work in Capability Myths Demolished and Robust Composition - Towards a Unified Approach to Access Control and Concurrency Control — it is what those frameworks become when keys cross machine boundaries. The Lowe/PKI Layer Cake - Kaminsky Patterson Sassaman-style critiques of X.509 PKI are simultaneously arguments for SPKI-style local-name systems: parser differentials and the trust-collapse-by-CA-misissuance problem largely vanish when certificates bind authority rather than identity.

In this vault

SPKI

Simple Public Key Infrastructure (Ellison et al. RFC 2693, 1999) — the IETF formalisation of the SPKI/SDSI design. Certificates bind keys to authorisations (5-tuples <Issuer, Subject, Delegation, Authority, Validity>) rather than to global identities. Authority delegation is first-class; chain discovery (Clarke et al. 2001) gives a sound and complete algorithm for deriving access decisions from collections of certificates. The cryptographic-identity counterpart of object-capability security in distributed systems.