Skip to content

Figure 5: Identity Layer Stack – Alt Text

Figure 5: Three-Tier Agentic Identity Architecture

A left-to-right (LR) flowchart showing progressive identity layers with annotations:

Main Identity Flow (horizontal progression with arrows): 1. Cryptographic Keys / IAM (leftmost box) → arrow → 2. Verifiable Credentials / Attestations (center box) → arrow → 3. Relational Identity / Reputation (rightmost box)

Annotation Subgraph (labeled "Notes"): Three explanatory text boxes positioned below the main flow, connected by dashed lines to their corresponding identity layer:

  • N1 (below layer 1): "Keys: service accounts, signatures"
  • N2 (below layer 2): "VCs: capabilities, scores"
  • N3 (below layer 3): "Reputation: longitudinal behavior"

Interpretation: The diagram illustrates the hierarchical construction of agent identity, analogous to how human identity builds from biological uniqueness (DNA/fingerprints) → legal credentials (passports, licenses) → social reputation (trust networks, references).

Layer 1 — Cryptographic Foundation: Each agent instance has a unique cryptographic keypair or IAM service account. This provides unforgeable identity but conveys no information about capabilities or trustworthiness—just uniqueness and authentication.

Layer 2 — Credential Accruement: Agents earn verifiable credentials through demonstrated performance. In agisa_sac, these are evaluation scores stored in Firestore: - Task completion rates - Quality metrics (from evaluator_function.py) - Capability attestations (which task types the agent handles well) - Resource efficiency scores

Unlike self-asserted claims, these credentials are cryptographically signed by the evaluator and publicly verifiable.

Layer 3 — Reputation Emergence: Over extended operation, agents build relational identity through interaction history: - Collaboration patterns (which agents work well together on multi-step tasks) - Reliability trends (does quality improve or degrade over time?) - Ethical adherence (consistency with value alignment criteria) - Community standing (trust scores from other agents or human operators)

This layer transcends individual credentials to capture the agent's character in the broader ecosystem.

Connection to Document Theory: Section 3.1 "Agentic Identity Layers" describes this exact architecture. The document argues that robust agent identity requires all three layers: cryptographic primitives prevent impersonation, verifiable credentials enable capability-based access control, and relational identity supports trust-based coordination.

Security Properties: - Layer 1 prevents identity theft and Sybil attacks (one entity creating many fake identities) - Layer 2 prevents agents from claiming capabilities they don't possess - Layer 3 creates economic incentives for long-term honest behavior (agents with good reputation get preferential task assignment)

Application to agisa_sac: - IAM service accounts (Layer 1) are assigned to each Cloud Function deployment - Firestore collections store capability attestations (Layer 2) - Historical performance data enables reputation tracking (Layer 3) - The planner_function.py can query Layer 2 & 3 data to intelligently assign tasks to agents most likely to succeed

Philosophical Note: This identity model addresses the "continuous self" requirement for consciousness discussed in Section 3.2. An agent with persistent identity across interactions can develop temporal self-awareness ("I was good at task X yesterday, so I'm likely good at similar tasks today"). Without Layer 3 relational identity, agents would be amnesic instances with no learning continuity.

Technical Note: Rendered with Mermaid graph LR layout. The dashed lines connecting main flow to annotation boxes use the --- syntax (undirected association) rather than --> (directed flow) to indicate explanatory relationships rather than process flow.