AEGIS AIAM-1 Specification Suite

Document: AIAM-1/Index (/identity/)
Version: 0.1 (Draft)
Part of: AEGIS Identity & Access Management for AI Agents
Last Updated: April 10, 2026


1. Overview

AIAM-1 (AEGIS Identity and Access Management, version 1) is a normative specification for identity and access management for AI agents (aIAM). It defines the identity, intent, authority, capability, delegation, session, attestation, revocation, interoperability, and threat model primitives required to govern autonomous and semi-autonomous AI agents as a distinct actor class.

AIAM-1 introduces Intent-Bound Access Control (IBAC), an authorization model in which every access decision is evaluated as a function of three inputs: identity, action, and intent context. IBAC is the first authorization model designed specifically for the agent actor class, where intent shifts dynamically within a single operational session.

1.1 The Problem

Identity and access management was built for two actor classes: humans and service accounts. AI agents are a third actor class that violates the assumptions underlying both:

Attempting to govern AI agents with existing IAM primitives results in agents that are either over-scoped (given broad credentials because narrow scoping would make them useless) or under-scoped (constrained to the point of being non-functional). Neither outcome is governance.

Scope caveat: AIAM-1 v0.1 is scoped to single-organization deployments. Cross-organization delegation — where principal chains cross legal and trust boundaries — is identified as an open problem in the Delegation, Threat Model, and Revocation chapters and is deferred to v0.2.

1.2 What AIAM-1 Defines

PrimitivePurposeSpecification Chapter
Composite IdentityFour-dimensional agent identity: model provenance, orchestration layer, goal context, principalIDENTITY
Intent ClaimsStructured assertions of purpose at the moment of actionINTENT
Intent-Bound Access Control (IBAC)Authorization as a function of (identity, action, intent)AUTHORITY
Capability ScopingTime-bounded, non-transitive, composition-governed capabilitiesCAPABILITIES
Delegation & Principal ChainsExplicit accountability chains with monotonic authority narrowingDELEGATION
Session GovernanceSessions as first-class governance boundariesSESSIONS
Attestation RecordsAction-level tamper-evident proof of governance decisionsATTESTATION
Revocation & Kill-SwitchPre-action revocation with propagation latency guaranteesREVOCATION
InteroperabilityMappings to OAuth 2.1, OIDC, SCIM, SAMLINTEROPERABILITY
Threat ModelSeven threat classes specific to agent IAMTHREAT MODEL
ConformanceRequirements checklist and conformance profilesCONFORMANCE

1.3 Out of Scope

AIAM-1 v0.1 does not define:


2. Relationship to Other Standards

2.1 Standalone Design

AIAM-1 is designed to be implemented independently of any specific governance runtime. Conformance to AIAM-1 does not require adoption of any other AEGIS specification. Cross-references to AEGIS artifacts in this specification are informative, not normative, unless explicitly stated.

2.2 AEGIS Ecosystem Integration

When deployed within an AEGIS-governed environment, AIAM-1 integrates with:

SpecificationIntegration PointRelationship
AGP-1 (Governance Protocol)AIAM-1 composite identity is the backing record for AGP-1 actor.id. AIAM-1 intent claims extend AGP-1 ACTION_PROPOSE with structured purpose assertions.AIAM-1 enriches AGP-1 identity and action semantics.
ATX-1 (Threat Matrix)AIAM-1 threat model (§3.10) cross-references ATX-1 tactics where applicable. ATX-1 TA007 (multi-agent collusion) and TA010 (semantic boundary violations) are directly relevant to AIAM-1 delegation and intent primitives.Complementary threat coverage.
ATM-1 (Threat Model)AIAM-1 threat classes overlap with ATM-1 attack vectors (AV-3.1 identity spoofing, AV-7.1 composition attacks, RC-2 credential revocation). AEGIS architecture provides baseline mitigations; AIAM-1 requires conformant implementations to address threats independently.Complementary defenses.
GFN-1 (Federation Network)GFN-1 DIDs identify federation nodes. AIAM-1 identity claims identify agents. These are structurally distinct — a federation node identity and an agent identity operate at different layers. AIAM-1 attestation records (action-level) complement GFN-1 attestations (node-level).Complementary identity at different layers.
RFC-0004 (Governance Event Model)AIAM-1 §3.10.4 requires that trust evaluations comply with the structural separation of security and reputation signals. Within AEGIS, this is normatively specified in RFC-0004 §5 (two-layer trust model).Architectural constraint.

2.3 External Standards

StandardRelationship
OAuth 2.1 / OIDCAIAM-1 identity claims map into OAuth 2.1 token formats. AIAM-1 adds intent and authority primitives that OAuth does not specify.
SCIMAIAM-1 extends SCIM schema for agent lifecycle management (provisioning, updating, deprovisioning).
XACML / ABACIBAC extends the ABAC evaluation model with intent context as a first-class input.
SAMLAIAM-1 supports SAML assertions for federated agent identity in environments that have not adopted OAuth 2.1.
NIST AI RMFAIAM-1 implements identity and access management controls required at the technical layer of an AI RMF program.
PBAC (Byun et al., 2005)IBAC extends Purpose-Based Access Control with structured intent claims, principal chains, and session governance.

3. Terminology

AIAM-1 introduces several terms that either extend or complement terminology used in other AEGIS specifications.

TermAIAM-1 DefinitionOther AEGIS Usage
PrincipalThe human, organization, or legal entity on whose behalf an agent acts. The accountable party.AGP-1/ATM-1 use “actor” for any entity submitting an action proposal (agent, human, or service). In AIAM-1, an agent is an actor; its principal is the entity accountable for the agent’s actions.
Intent claimA structured assertion of the purpose of a specific action at the moment the agent proposes it.Not present in AGP-1. AGP-1 treats intent as implicit in ACTION_PROPOSE parameters. AIAM-1 intent claims extend, not replace, AGP-1 message structures.
SessionA bounded governance context defined by goal, time window, capability envelope, and accountability chain.AGP-1/RFC-0002 use session_id as a stateless correlation key. AIAM-1 sessions are stateful governance boundaries.
AttestationA cryptographically signed record of an action taken by an agent, including identity, intent, authority decision, capabilities, outcome, and principal chain.GFN-1 uses governance.attestation.v1 for node-level governance posture statements. AIAM-1 attestation is action-level. Both coexist as complementary records at different granularities.
IBACIntent-Bound Access Control. Authorization evaluated as a function of (identity, action, intent context).Not present in other AEGIS specs. RFC-0003 uses role-based capability scoping, which is a special case of IBAC where intent context is wildcard.
Composite identityFour-dimensional agent identity: model provenance, orchestration layer, goal context, principal.AGP-1 uses actor.id (opaque string) and actor.type (enum: agent, user, service). AIAM-1 composite identity is the rich backing record that an AGP-1 actor.id resolves to.

4. Authorization Model Landscape

Authorization models have evolved to match the complexity of the actors they govern. Each model was a response to a new actor class or deployment pattern:

ModelEraDecision InputsCore QuestionAgent Suitability
DAC (Discretionary)1970sObject owner grantsWho owns this?Insufficient: agents do not own resources; ownership is not a meaningful governance dimension.
MAC (Mandatory)1970sSecurity labelsWhat clearance do you have?Insufficient: static clearance levels cannot govern dynamic, goal-shifting actors.
RBAC (Role-Based)1990sIdentity + role membershipWhat role are you in?Insufficient: agents shift goals within roles; roles are too coarse for action-level governance.
ABAC (Attribute-Based)2000sActor attributes + resource attributes + environmentWhat do you look like?Partial: attributes can encode some context, but intent is not a first-class attribute.
PBAC (Purpose-Based)2005Identity + declared purposeWhy do you need this?Partial: captures purpose but lacks principal chains, session governance, and structured intent validation.
IBAC (Intent-Bound)2026Identity + action + intent contextWhy are you doing this, right now, in this context?Designed for agents: intent is structured, validated against goal context, and preserved in attestation records. Generalizes RBAC, ABAC, and PBAC.

IBAC strictly generalizes prior models: an RBAC policy is an IBAC triple where the identity pattern matches on role, the action pattern is wildcard-scoped, and the intent context pattern is wildcard. An ABAC policy is an IBAC triple where attributes are mapped to identity and action patterns with wildcard intent. Organizations migrating from RBAC or ABAC can adopt IBAC incrementally by starting with wildcard intent patterns and progressively adding intent constraints as they mature.


5. Document Suite

5.1 Normative Documents

DocumentStatusDescription
AEGIS_AIAM1_INDEX.mdDraftThis document. Suite index and overview.
AEGIS_AIAM1_IDENTITY.mdDraftComposite identity model, cryptographic attestation, JSON schema.
AEGIS_AIAM1_INTENT.mdDraftIntent claims as structured assertions, validation, spoofing detection.
AEGIS_AIAM1_AUTHORITY.mdDraftIntent-Bound Access Control (IBAC), authority binding triples, policy format.
AEGIS_AIAM1_CAPABILITIES.mdDraftCapability scoping, composition governance, non-transitivity, revocation.
AEGIS_AIAM1_DELEGATION.mdDraftPrincipal chains, monotonic authority narrowing, delegation depth limits.
AEGIS_AIAM1_SESSIONS.mdDraftSession as governance boundary, four-dimensional bounding.
AEGIS_AIAM1_ATTESTATION.mdDraftAttestation records, tamper-evidence, retention.
AEGIS_AIAM1_REVOCATION.mdDraftKill-switch, propagation latency, pre-action enforcement.
AEGIS_AIAM1_INTEROPERABILITY.mdDraftOAuth 2.1, OIDC, SCIM, SAML mappings.
AEGIS_AIAM1_THREAT_MODEL.mdDraftSeven threat classes, ATX-1 cross-references, residual risk.
AEGIS_AIAM1_CONFORMANCE.mdDraftConformance checklist, profiles, conformance statements.

5.2 Schemas

Ownership note: The AIAM schema set currently lives in aegis-governance/aegis-core/schemas/aiam/ as a governance-owned extension. Shared cross-repository contracts for the broader AEGIS ecosystem are canonically owned by aegis/schemas/. AIAM schemas should be promoted into the canonical shared schema set only after the specification stabilizes enough for downstream product and SDK consumption.

SchemaLocationDescription
common.schema.jsonaegis-core/schemas/aiam/Shared type definitions: ID formats, timestamps, signatures, hashes. All other schemas $ref into this.
identity_claim.schema.jsonaegis-core/schemas/aiam/JSON Schema for AIAM-1 composite identity claims.
intent_claim.schema.jsonaegis-core/schemas/aiam/JSON Schema for AIAM-1 intent claims.
attestation_record.schema.jsonaegis-core/schemas/aiam/JSON Schema for AIAM-1 attestation records.
delegation_record.schema.jsonaegis-core/schemas/aiam/JSON Schema for AIAM-1 delegation records.

5.3 Position Paper

DocumentLocationDescription
AIAM-1-position-paper-v0.1.mddocs/position-papers/aiam/Position paper establishing the normative scope of AIAM-1 v0.1.

6. Normative Language

This specification uses the terms MUST, MUST NOT, SHOULD, SHOULD NOT, and MAY as defined in RFC 2119. A conformant AIAM-1 implementation satisfies all MUST and MUST NOT requirements across all normative documents in this suite.


7. Version History

VersionDateChanges
0.12026-04-10Initial draft. Position paper and specification suite.

AEGIS™ | “Capability without constraint is not intelligence”™
AEGIS Initiative — AEGIS Operations LLC