Papers
Topics
Authors
Recent
Search
2000 character limit reached

AI Identity: Standards, Gaps, and Research Directions for AI Agents

Published 25 Apr 2026 in cs.AI and cs.CR | (2604.23280v1)

Abstract: AI agents are now running real transactions, workflows, and sub-agent chains across organizational boundaries without continuous human supervision. This creates a problem no current infrastructure is equipped to solve: how do you identify, verify, and hold accountable an entity with no body, no persistent memory, and no legal standing? We define AI Identity as the continuous relationship between what an AI agent is declared to be and what it is observed to do, bounded by the confidence that those two things correspond at any given moment. Through a structured survey of industry trends, emerging standards, and technical literature, we conduct a gap analysis across the full agent identity lifecycle and make three contributions: (1) a structural comparison of human and AI identity across four dimensions (substrate, persistence, verifiability, and legal standing) showing that the asymmetry is fundamental and that extending human frameworks to agents without structural modification produces systematic failures; (2) an evaluation of current technical and regulatory documents against the identity requirements of autonomous agents, finding that none adequately address the challenge of governing nondeterministic, boundary-crossing entities; and (3) identification of five critical gaps (semantic intent verification, recursive delegation accountability, agent identity integrity, governance opacity and enforcement, and operational sustainability) that no current technology or regulatory instrument resolves. These gaps are structural; more engineering effort alone will not close them. Foundational research on AI identity is the central conclusion of this report.

Summary

  • The paper establishes fundamental asymmetries between human and AI identities, analyzing dimensions such as substrate, persistence, verifiability, and legal standing.
  • It quantifies the prevalence of non-human identities (over 140:1 in enterprises) while highlighting fragmented market solutions and regulatory inconsistencies in agent governance.
  • It advocates shifting from static credential checks to continuous, confidence-bounded identity frameworks to tackle semantic intent verification and opaque delegation chains.

Authoritative Summary of "AI Identity: Standards, Gaps, and Research Directions for AI Agents" (2604.23280)

Structural Differences Between Human and AI Identity

The paper establishes a foundational analysis of the asymmetries between human and AI agent identity, spanning four primary dimensions: substrate, persistence, verifiability, and legal standing. Human identity is anchored in biological substrates (e.g., DNA, neural tissue), conferring lifetime persistence and verifiability through biometric signals and social institutions. Legal frameworks grant rights and obligations directly to humans.

Conversely, AI agents—modeled as non-human identities (NHI)—lack any biological substrate, exhibit ephemeral persistence heavily sensitive to configuration and runtime context, and are fundamentally nondeterministic. Verifiability is limited: two identical model deployments may diverge in behavior due to stochastic computation or mutable context. Legal standing is absent, often reconstructed through delegated authorization chains linked to humans or organizations.

The report dissects NHI into model, agent, workload, and delegated types, each with distinct life cycles and practical limitations. Model-level identity refers to the artifact (weights, architecture, provenance), agent-level to runtime configuration and persona, workload-level to ephemeral execution contexts, and delegated-level to externally granted permissions. None mirrors the permanence or accountability of human identity systems.

Critical Analysis of Market, Standards, and Regulatory Landscape

Market solutions have begun to address agent enrollment, lifecycle management, and runtime credentialing but remain fragmented. Products such as Saviynt, Astrix, HashiCorp Vault, and AgenticTrust operate in discrete silos, with varied trust roots and incompatible credential semantics.

Standards efforts are equally divided. SPIFFE/WIMSE provides ephemeral workload attestation; OAuth 2.0 supports limited synchronous delegation; and OpenID Agentic AI and MCP clarify agent-environment boundaries, yet none span the full lifecycle, especially multi-hop delegation and cross-domain authorization. Regulatory frameworks across the EU, US, China, Japan, and Singapore are characterized by heterogeneity and at times conflicting requirements, targeting different aspects of AI identity and content attribution.

The paper demonstrates that authentication of containers (e.g., tokens, certificates, SVIDs) confers no guarantees about behavioral consistency or semantic intent, especially under nondeterminism and context drift. Authorization protocols lack enforceable scope attenuation in delegation chains. Credentials, DID/VC solutions, and ZKP deployments promise selective disclosure and supply-chain traceability but do not resolve provenance or accountability at the required granularity. Audit and attestation mechanisms (TEE, TPM, immutable logs, SVIP) ensure code and environment integrity but not genuine behavioral conformance.

Identification of Structural Gaps

Five unresolved, structural gaps are identified:

  • Semantic Intent Verification: No current infrastructure can verify that agent behavior aligns with the genuine intentions of its principal, nor detect prompt injection or reasoning hijacks.
  • Recursive Delegation Accountability: Multi-hop delegation chains are fundamentally opaque; current standards cannot cryptographically or operationally trace authorization provenance beyond the first hop.
  • Agent Identity Integrity: Cloning, puppeteering, credential sharing, and Sybil attacks exploit the lack of stable substrate and instance uniqueness. Hardware attestation and anomaly detection offer partial mitigation.
  • Governance Opacity and Enforcement Paradox: Enforcement regimes tend to exclude agents lacking enterprise credentials, leading to shadow agent deployments outside monitoring and audit frameworks, and reproduce credential inequalities across resource-divided organizations.
  • Operational Sustainability: The energy and computational cost of cryptographic verification and attestation at planetary scale has not been assessed; scalability and ecological viability of current approaches are unresolved.

The paper provides strong numerical evidence: NHIs outnumber human identities in enterprise settings by over 140:1, yet visibility and governance coverage drastically lag, with only about half of agent deployments actively monitored.

Theoretical and Practical Implications

Practically, these gaps preclude reliable accountability or liability assignment for AI agents operating across organizational boundaries. The inability to verify intent, track delegation chains, or prevent mass impersonation undermines the deployment of AI agents in critical workflows, such as finance and healthcare.

Theoretically, the paper advocates for a shift from binary credential identity to a continuous correspondence model, wherein AI identity is represented probabilistically as the ongoing relationship between declared attributes and observed behaviors, bounded by dynamically updated confidence scores. This advances the discourse beyond credential checking to dynamic and adaptive identity reckoning, suitable for agent-centric environments.

Emergent research priorities include semantic intent attestation, robust scope attenuation, instance-level identity binding, tiered verification proportional to risk, and operational cost modeling for cryptographic infrastructure.

Speculation on Future Developments

Future advances are likely to integrate human-in-the-loop attestation, intent-aware behavioral contracts, and immutable, cross-organizational delegation logging. Approaches may combine deep ZKP frameworks, adaptive anomaly detection, and hardware-rooted identity substrates. Regulatory harmonization and flexible onboarding for resource-constrained agents will be crucial for equitable and scalable agent ecosystems. Sustainability concerns will drive research into verification batching and ecological cost accounting.

Conclusion

The paper rigorously demonstrates that current identity, credentialing, authorization, and governance paradigms—architected for humans and deterministic machines—are structurally inadequate for autonomous, nondeterministic AI agents. The field must ground AI identity in continuous, confidence-bounded correspondence frameworks and address the five identified gaps through foundational interdisciplinary research. These are not just engineering challenges; they delimit the architecture of future agent ecosystems and the boundaries of accountability and liability in AI deployment.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Collections

Sign up for free to add this paper to one or more collections.

Tweets

Sign up for free to view the 1 tweet with 0 likes about this paper.