notes: research/papers/hardman-the-shape-of-delegated-authority

This data as json

path title type status classification words body
research/papers/hardman-the-shape-of-delegated-authority The Shape of Delegated Authority summary v0.1 — 2026-10-15: drafted by Claude from the HTML page (version 1.2); unread by Mark; marked partner on Mark's word 2026-10-21, paraphrase only partner 587 # The Shape of Delegated Authority **Abstract.** A 2026 paper by Daniel Hardman proposing a model of delegated authority that makes the category of a delegate's act machine-checkable and lets a stranger verify authority without consulting its issuer. It argues that access control, action taxonomies and decentralized identity each keep only part of what the law of agency long held together, and it draws on KERI (Key Event Receipt Infrastructure) and ACDC (Authentic Chained Data Container) mechanisms for the identity and verification side. It is for architects of authorization for human and AI agents. ## Summary The paper opens with the contrast between handing a delegate a bounded key and handing over a signet ring, and argues that digital systems, through API keys and bearer tokens, mostly hand out rings. Agentic AI makes the gap urgent: authenticating an agent says who is at the door, not what it may do or for whom. Section 2 argues that three research literatures each reduced the old practice of agency: formal access control and capability tokens (rigorous on acts, silent on the relationship), taxonomies of action (rich on acts, single-actor), and the self-sovereign identity work on delegation and guardianship (rich on the relationship, ad hoc on acts). Section 3 explains why an intent grammar such as the author's Syntelos taxonomy is one axis but not enough, since it excludes acts on tools and is two-party. Section 4 gives the act a shape: telos (purpose), effect (observe, create, modify, preserve, destroy) and a named kind of state (information, record, commitment, authority, resource, relationship), modulated by the target's stakes and composed as vectors. The authorization gate and its owner are derived from these coordinates. Section 5 adds a relationship facet: principal, relationship type (delegation, guardianship, controllership, stewardship), liable party, presented identity, duties and revocation, and separates authority to act from authority to authorize. Section 6 makes identity facet-granular, with one cryptographic identifier per who, role and context triple. Section 7 sets open-loop verification (credentials checked from key state alone) against closed-loop capabilities, and notes constraints enforced by construction in cooperative delegation. Section 8 lists what is settled and what stays open, including the root of the chain (with joint issuance from the verifiable-dossier work as a candidate), attenuation over long chains, aggregation, and standing authority. ## Key terms delegated authority, telos, effect, state-kind, authorization gate, intent boundary, stewardship, liable party, facet, open-loop verification, authority to authorize ## Claims and their evidence - API keys and bearer tokens grant open-ended authority that makes the holder indistinguishable from the account. [informative] - Access control, action taxonomies and decentralized identity each drop either the full act surface or the "on behalf of, in whose interest, who is liable" dimension. [informative] - Authorization gates can be derived from an act's telos, effect, state-kind and stakes rather than hand-maintained. [informative] - Verification of authority must succeed from the credential and key state alone; a registry consulted at verification time is a closed loop. [informative] - Joint issuance refines KERI multisignature into a lighter, composable threshold mechanism. [unverified] - The closed sets (five effects, six state-kinds, four relationship types) are the fewest that suffice; the paper states this is claimed, not proved. [unverified] ## Related Syntelos: A Hierarchical Taxonomy of Intent in Digital Interactions; Intent and Boundaries: A Framework for Digital Agency; Identity Facets; Opaque Identifier Aliases; Verifiable Dossiers; KERI specification; ACDC specification ## Provenance Drafted by Claude on 2026-10-15 from the HTML page at dhh1128.github.io/papers/sda.html, read in full (sections 1 to 8). Not yet read by Mark.

Links from other tables