Pith. sign in

REVIEW 5 major objections 5 minor 2 cited by

Beyond DNS: Unlocking the Internet of AI Agents via the NANDA Index and Verified AgentFacts

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper proposes a lean NANDA index paired with signed AgentFacts documents as the discovery and trust layer for an Internet of AI agents, claiming sub-second resolution and revocation at trillion-agent scale.

desk verdict Useful architecture proposal whose 'concrete guarantees' are really design goals, with a load-bearing gap in sub-second revocation. read the letter →

arxiv 2507.14263 v1 pith:TSQTZ43X submitted 2025-07-18 cs.NI cs.AIcs.CRcs.MA

Ramesh Raskar , Pradyumna Chari , John Zinky , Mahesh Lambe , Jared James Grogan , Sichao Wang , Rajesh Ranjan , Rekha Singhal
show 10 more authors
Shailja Gupta Robert Lincourt Raghu Bala Aditi Joshi Abhishek Singh Ayush Chopra Dimitris Stripelis Bhuwan B Sumit Kumar Maria Gorskikh
This is my paper
classification cs.NIcs.AIcs.CRcs.MA
keywords AIagentdiscoveryNANDAindexFactsschemaVerifiableCredentialsdecentralizedidentifiersinteroperabilityprivacy-preservingresolutionDNSlimitations
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper is a design proposal for a new discovery layer for the coming Internet of AI agents. It argues that DNS, built for static human-facing endpoints, cannot handle billions to trillions of agents that appear, move, change capabilities, and need revocation in seconds. The proposed answer is a lean index of small signed records that point not to endpoints but to dynamic, cryptographically signed metadata documents called AgentFacts, which carry endpoints, capabilities, audits, and telemetry. The claim is that this separation makes resolution fast (under one second of global propagation), private, and verifiable while keeping the index itself nearly static. If the architecture holds, agents could be discovered, authenticated, and routed without rewriting DNS or trusting self-advertised claims.

What carries the argument

The load-bearing object is the AgentAddr record: a signed, cacheable pointer of about 120 bytes returned by the index, carrying an agent ID, a human-readable URN, a public facts URL, an optional private facts URL, an optional adaptive resolver URL, a TTL, and an Ed25519 signature. It does the work of a DNS record but deliberately points at metadata rather than at an endpoint, so all volatile state lives elsewhere. The AgentFacts document is the second mechanism: a self-describing JSON-LD file signed as a Verifiable Credential that lists endpoints, capabilities, skills, evaluations, and telemetry, hosted on an agent-owned domain, IPFS, or a third-party host. Together they enforce the paper's core separation of static identity resolution from dynamic metadata and runtime routing, which is what converts a fully connected N-by-N agent network into a simpler 2N handshake problem.

What would settle it

Deploy the index on a globally distributed testbed, spawn and revoke agents at churn rates approaching the claimed 10,000 updates per second per shard, and measure the p99 time from AgentAddr publication to consistent global resolution and from status-list revocation to rejection; if either exceeds one second under realistic network conditions, the paper's central latency and revocation guarantees fail.

Watch

Extended reading notes

Core claim

The paper's central claim is that the bottleneck for an agentic internet is not raw lookup volume but the coupling of stable identity to volatile metadata: DNS maps names to fixed endpoints and updates slowly, so every agent move or revocation forces an expensive global write. NANDA breaks that coupling with three layers: a lean index that stores signed, cacheable AgentAddr records (about 120 bytes) mapping an agent name to pointers; AgentFacts, JSON-LD documents signed as standard Verifiable Credentials, which hold endpoint lists, capability assertions, certifications, and telemetry; and an adaptive resolution layer that selects static (1-6 hour), rotating (5-15 minute), or ephemeral (30-60 second) endpoints. Because AgentFacts are hosted at self-selected locations and can be updated without index writes, the index absorbs only a small, stable write load, and short-lived credentials checked against a status list give sub-second revocation. Privacy comes from a dual-path lookup: a public facts URL served by the agent and an optional private facts URL served by third-party or decentralized storage, so a requester can fetch metadata without revealing itself to the agent's operator. The design is explicitly a quilt: NANDA-native agents, enterprise registries, tool-server registries, existing agent-card formats, and Web3-style decentralized identifiers can all be visible through the same interface.

Load-bearing premise

The load-bearing premise is that the Internet will host billions to trillions of autonomous AI agents whose coordination needs cannot be met by DNS; if agent populations stay modest, or if extended DNS records and TTL tuning prove sufficient, the motivation for a new global index collapses.

Editorial extensions

If this is right

  • Newly spawned agents become globally resolvable in under a second, because an AgentAddr write to the index is the only global step and all richer metadata follows from signed, cacheable facts.
  • Endpoint churn no longer causes index write storms: blue/green deploys, geo-failover, and DDoS shuffling happen on rotating or adaptive TTLs, while the index record itself stays unchanged for hours.
  • A client can discover and vet an agent without revealing its own identity or even contacting the agent's infrastructure, by resolving through a private facts URL hosted on IPFS or a third party.
  • Capability claims become auditable rather than self-advertised, because skills, certifications, and evaluations are signed by issuers in federated trust zones, and clients can require whitelisted issuers or threshold verification.
  • The index can federate enterprise registries, tool-server registries, existing agent-card agents, and decentralized-identifier identities into one quilt, so adoption does not require abandoning existing agent platforms.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Inference: the architecture's advantage over DNS should scale with metadata churn rate, not with raw agent count; a meaningful benchmark would compare end-to-end resolution latency under increasing spawn and revoke frequency against a DNS baseline with equivalent TTL tuning.
  • Inference: the AgentFacts schema already contains the fields (latency budgets, availability, audit trails, performance scores) needed to double as a machine-readable service-level agreement, which could let reputation scoring emerge directly from verification rather than from separate rating systems.
  • Inference: the privacy path as specified shields the requester from the agent operator but not from the third-party private facts URL host; fully least-disclosure discovery would require the zero-knowledge assertions listed as future work or anonymous hosting.
  • Inference: the sub-second revocation guarantee rests on status-list infrastructure that the paper does not analyze; a deployment test measuring revocation propagation under load would be the quickest way to separate the architecture's structural claims from its timing targets.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. This paper proposes the NANDA index, a layered naming and discovery architecture intended for a future 'Internet of AI agents.' The design separates a lean, signed index record (AgentAddr) from JSON-LD-based AgentFacts documents that carry endpoints, capabilities, and W3C Verifiable Credentials, with TTL-based caching, adaptive resolvers, and a privacy-preserving private facts path. The abstract claims that this architecture delivers five concrete guarantees: quilt-like registry interoperability, rapid global resolution for newly spawned agents, sub-second revocation and key rotation, schema-validated capability assertions, and privacy-preserving least-disclosure discovery. The paper is labeled v0.3 'Work in Progress, Request for Comments' and presents design goals, an example schema, resolution flows, deployment models, and a list of open questions.

Significance. The paper is a design/vision document rather than a validated systems paper. If its guarantees were substantiated, NANDA could serve as a useful reference architecture for agent naming and metadata, particularly because it builds on existing standards (W3C VCs, DIDs, JSON-LD) and provides a concrete AgentFacts schema in the appendix. The comparison with Google's A2A Agent Card in Table 4 and the candid enumeration of open questions in Section XI are useful contributions. On the other hand, the manuscript contains no derivation, simulation, prototype evaluation, or measurements supporting the stated quantitative guarantees; the 'five concrete guarantees' are asserted as design outcomes. The significance is therefore conditional on future specification work, not on the present evidence.

major comments (5)
  1. [Abstract and Section I] The abstract and Section I state that 'our architecture design delivers five concrete guarantees,' but the paper provides no derivation, simulation, prototype, or measurement for any of the five. Table 2's green ticks denote 'fully specified' design features, not validated properties. Since these guarantees are the paper's central claim, the claim is load-bearing: either an evaluation must be added or the guarantees must be reclassified as design goals or hypotheses.
  2. [Section I, V.E, VII.D, X.2] The sub-second revocation guarantee is not supported. Section I argues that 'short-lived verifiable credentials (<5 min) allow sub-second revocation,' but a 5-minute validity window bounds revocation delay by about 5 minutes unless every verification performs an uncached online VC-Status-List check on every resolution. Section VII.D only states that revocation lists are 'queried by clients during interaction,' while Section V.E deliberately caches AgentAddr records with TTLs of 1-6 hours and caches AgentFacts, so a cached record can outlive a revocation. Section X.2 defers formalizing revocation protocols to future work. Because revocation is guarantee (3), the architecture's central value proposition is unverified and possibly inconsistent with its own caching design.
  3. [Section I and Section II] The core problem statement motivating the paper—that DNS cannot support trillions of agents and sub-second global propagation—is asserted without independent evidence and is cited to the project's own companion paper [7]. This is circular as a justification for the architecture. The authors should either provide external evidence for the trillion-agent premise or explicitly frame the contribution as conditional on that premise.
  4. [Abstract vs. Sections II-VIII] The abstract states 'We formalize the AgentFacts schema, specify a CRDT-based update protocol, and prototype adaptive resolvers,' but the body contains no specification of a CRDT-based update protocol and no description or evaluation of an adaptive resolver prototype. Section II.D describes asynchronous metadata publishing but does not present a CRDT design. The contribution list and the body must be reconciled: either add the CRDT specification and prototype details or revise the abstract.
  5. [Section II.A and Table 2] The claim of a '~10^4×' reduction in index writes relative to DNS is stated without a model or derivation. The factor appears to depend on assumptions about update rates and record churn that are not given. A quantitative claim of this kind should be derived from an explicit model or replaced with a testable prediction parameterized by the actual update and query workloads.
minor comments (5)
  1. [Table 4 and Section V.A] The passage following Table 4 says 'The remainder of Section IV specifies how those features are encoded in JSON-LD,' but the AgentFacts schema is specified in Section V, not Section IV.
  2. [Table 4 and Section I] The takeaway under Table 4 says 'millisecond revocation via VC-Status lists,' while Section I and Table 4's row state 'sub-second revocation' or '<1 s.' These numbers should be made consistent.
  3. [Section VI.B and VI.C] The component name is spelled inconsistently as 'AdaptiveResolver' and 'Adaptiveresolver' in different places; a single spelling should be used throughout.
  4. [Reference [17]] The Goldwasser-Micali-Rackoff reference lists the SIAM Journal on Computing with volume 18, issue 1, pages 186-208, but dates it as 1985; the journal version appeared in 1989, and the 1985 date corresponds to the STOC proceedings.
  5. [Appendix] The example schema field 'jurisdiction': “USA” uses typographic curly quotes instead of straight quotes, which will break JSON parsing if the example is directly used.

Circularity Check

1 steps flagged · score 4.0 of 10

The motivating problem statement is sourced to the project's own companion paper via [7], and the five 'concrete guarantees' are restatements of design goals rather than independently derived results; however, the underlying mechanisms rely on external standards, so the architecture itself is not tautological.

  1. self citation load bearing [Section II 'Problems we are solving [7]'; Section I 'cannot be realized within the constraints of DNS']
    "Problems we are solving [7]: ... We argue that the emergence of an open, agentic internet cannot be realized within the constraints of DNS."

    The paper's central premise—that DNS is an insurmountable bottleneck for a trillion-agent internet—is not demonstrated in this paper; it is attributed to [7], a companion paper from the same Project NANDA. The architecture is then presented as solving exactly those cited problems ('For each problem [7], ask what single property...'). Thus the need for the NANDA index rests on a self-citation that is not independently verified in this draft. This is load-bearing for the paper's motivation and for the framing of the five 'concrete guarantees' as solutions. The technical mechanisms (W3C VCs, DIDs, IPFS, JSON-LD) are external standards, so the core architecture is not itself a tautology; the circularity is confined to the problem statement and the guarantee framing.

full rationale

This draft contains no equations, fitted parameters, or first-principles derivations; its 'five concrete guarantees' are presented in the abstract as delivered results but in Section II are explicitly design goals forming an 'orthogonal checklist,' with Table 2 marking several as partial or open. The one genuine circular step is the load-bearing self-citation: the problems that motivate the architecture are cited to the project's own companion paper [7], and the architecture is then claimed to solve those same self-referential problems. The sub-second revocation and <1s global propagation claims are asserted rather than derived, and the paper's own TTL caching (1-6 h for static endpoints) makes sub-second revocation questionable, but that is an unsupported correctness risk, not a circular reduction. Because the substantive building blocks (W3C Verifiable Credentials, DIDs, JSON-LD, IPFS, VC-Status-List) are external and the architectural content has independent substance, the circularity is partial rather than total, warranting a score of 4 rather than higher.

Assumptions & free parameters 5 free parameters · 4 assumptions · 5 invented entities

The ledger shows that the central claims rest on unmeasured design parameters, domain assumptions about agent-scale demand, and newly invented components with no independent evidence. The most important entries are the demand assumption and the unvalidated revocation mechanism.

free parameters (5)
  • AgentAddr record size limit (<=120 bytes) = 120 bytes
    Chosen by hand in Section II.A and IV to argue for reduced index write load; no measurement of real AgentAddr sizes, and the Table 3 example with URLs and signature exceeds 120 bytes by inspection.
  • TTL tier durations = static 1-6h, rotating 5-15min, adaptive 30-60s
    Proposed in Section V.E and VI without empirical tuning or workload analysis; these values are illustrative.
  • Index write reduction factor vs DNS (10^4x) = 10^4
    Claimed in Section II.A and III to motivate scalability; no derivation is provided for the factor.
  • Global propagation target (<1s) = <1s
    Stated as a design target in the abstract and Section I; no simulation or deployment evidence is given.
  • Privacy-path latency overhead (30-60ms) = 30-60ms
    Estimated in Table 2 as added latency for the PrivateFactsURL path; no measurements support the range.
assumptions (4)
  • domain assumption The internet will host billions to trillions of autonomous AI agents requiring millisecond-scale coordination.
    Section I asserts this demand scenario; no evidence or citation is provided beyond the authors' own companion paper [7].
  • domain assumption DNS update cycles and ownership-only trust cannot meet agent needs and cannot be extended to do so.
    Section I states this as the motivation for a new index; the paper does not evaluate proposals to extend DNS (e.g., DNS-based service discovery, ECS, or shorter TTL regimes).
  • domain assumption W3C Verifiable Credentials and VC-Status-List revocation can provide sub-second revocation globally.
    Section VII.D references VC-Status-List but does not analyze propagation latency, availability, or the time to detect a compromised credential.
  • domain assumption A federation of independent registries (the quilt model) can interoperate with consistent metadata standards and trust frameworks.
    Section IV.B describes the quilt model but leaves cross-registry interoperability protocols as co-dependencies, not specified details.
invented entities (5)
  • NANDA index
    purpose: Global, lean discovery layer for AI agents, implemented as a quilt of registries.
    No implementation, deployment, or simulation is provided; the index exists only as a design in Sections II-IV.
  • AgentAddr
    purpose: Signed, cacheable address record binding agent name to metadata and routing pointers.
    Defined in Section IV with an example in Table 3, but no reference implementation exists.
  • AgentFacts
    purpose: Self-describing, verifiable JSON-LD metadata document for agent capabilities, endpoints, and credentials.
    A schema is sketched in the appendix as a commented JSON example, but no formal JSON Schema validation or tooling is shipped.
  • PrivateFactsURL dual-path resolution
    purpose: Hide requester identity by resolving metadata through third-party or decentralized hosting.
    Section VII.C describes the mechanism and relies on IPFS/Tor relays, but no implementation or latency measurement is provided.
  • AdaptiveResolver
    purpose: Programmable routing component for load balancing, geo-dispatching, and DDoS mitigation.
    Section VI describes the component and possible implementations (Envoy, NGINX, serverless), but no prototype is described despite the abstract claiming one.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond DNS: Unlocking the Internet of AI Agents via the NANDA Index and Verified AgentFacts." pith.science (2026). https://pith.science/paper/TSQTZ43X

@misc{pith2026250714263,
  author       = {Pith},
  title        = {Pith review of: Beyond DNS: Unlocking the Internet of AI Agents via the NANDA Index and Verified AgentFacts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TSQTZ43X}},
  note         = {Machine review of arXiv:2507.14263}
}
read the original abstract

The Internet is poised to host billions to trillions of autonomous AI agents that negotiate, delegate, and migrate in milliseconds and workloads that will strain DNS-centred identity and discovery. In this paper, we describe the NANDA index architecture, which we envision as a means for discoverability, identifiability and authentication in the internet of AI agents. We present an architecture where a minimal lean index resolves to dynamic, cryptographically verifiable AgentFacts that supports multi-endpoint routing, load balancing, privacy-preserving access, and credentialed capability assertions. Our architecture design delivers five concrete guarantees: (1) A quilt-like index proposal that supports both NANDA-native agents as well as third party agents being discoverable via the index, (2) rapid global resolution for newly spawned AI agents, (3) sub-second revocation and key rotation, (4) schema-validated capability assertions, and (5) privacy-preserving discovery across organisational boundaries via verifiable, least-disclosure queries. We formalize the AgentFacts schema, specify a CRDT-based update protocol, and prototype adaptive resolvers. The result is a lightweight, horizontally scalable foundation that unlocks secure, trust-aware collaboration for the next generation of the Internet of AI agents, without abandoning existing web infrastructure.

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. AI Tool Discovery at Scale: All You Need is DNS

    cs.AI 2026-04 reject novelty 5.0 of 10

    ToolDNS repurposes DNS names and EDNS0 payloads to perform hierarchical semantic tool discovery with O(log N) name resolutions and a reported 95.26% search-space reduction.

  2. AI Agents with Decentralized Identifiers and Verifiable Credentials

    cs.CR 2025-10 conditional novelty 4.0 of 10

    A prototype shows AI agents can authenticate using ledger-anchored W3C DIDs and verifiable credentials, but LLM-orchestrated security steps fail often and once skipped authentication.

Reference graph

Works this paper leans on

35 extracted references · 30 canonical work pages · cited by 2 Pith papers

  1. [7]

    Connection to the endpoint is authenticated (e.g., via OAuth2 or JWT) and established

    Endpoint Resolution & Connection ● If adaptive_Resolver_url is used, the client sends a request and receives either: A redirect to the optimal endpoint or A signed ephemeral token. Connection to the endpoint is authenticated (e.g., via OAuth2 or JWT) and established. Clients may choose the resolution strategy based on Latency and load-balancing needs, pri...

    work page

  2. [1]

    AgentAddr (≤ 120 B) containing TTL and pointers (FactsURL, PrivateFactsURL, AdaptiveResolverURL)

    Index Level (Anchor Tier) : ○ Provides decentralized mapping from agent identifiers (IDs) to metadata URLs i.e. AgentAddr (≤ 120 B) containing TTL and pointers (FactsURL, PrivateFactsURL, AdaptiveResolverURL). ○ Records are cacheable and Ed25519-signed, slashing index-write overheads while blocking tampering. Maintains only essential, static metadata such...

    work page

  3. [2]

    ○ Enables frequent, independent updates without requiring index-level intervention

    AgentFacts Level (Metadata Distribution Tier): ○ Hosts self-describing JSON-LD documents (AgentFacts), each signed as a W3C VC., that contain endpoint lists, capability descriptors, telemetry configurations, authentication protocols, and credentialed evaluations. ○ Enables frequent, independent updates without requiring index-level intervention. V0.3, Wor...

    work page

  4. [3]

    ○ Includes support for decentralized adaptive resolvers that can load-balance, geolocate, or behavior-route traffic as needed

    Dynamic Resolution Level (Adaptive Routing Tier): ○ Dynamically interprets AgentFacts metadata to resolve live endpoints (TTL 1-6 h), rotating pool (TTL 5-15 min), or apply adaptive routing policies, and maintain privacy constraints. ○ Includes support for decentralized adaptive resolvers that can load-balance, geolocate, or behavior-route traffic as need...

    work page

  5. [4]

    Translation Assistant v1.2.1

    Adaptive Routing Resolution: When the agent supports dynamic, context-aware routing (e.g., for load balancing, geo-awareness, or DDoS protection), the index returns the resolver URL directly in the AgentAddr. The AdaptiveResolver may return temporary signed endpoints or session tokens. The AgentFacts is not required unless client policies demand additiona...

    work page 2022

  6. [5]

    The AgentFacts is verified using digital signatures and credential chains

    Metadata Resolution (Optional) ○ If trust evaluation or capability validation is needed, the client fetches the AgentFacts from either: facts_url (direct access) or private_facts_url (privacy-preserving access). The AgentFacts is verified using digital signatures and credential chains

    work page

  7. [6]

    Endpoint Discovery ○ Clients inspect AgentFacts metadata (if fetched) or AgentAddr (if endpoint is included directly) for: ● Static Endpoints : Stable communication URIs ● Rotating Endpoints : Short-lived, dynamic URIs ● Adaptiveresolver URL : A programmable routing microservice

    work page

  8. [8]

    This defines how long the record may be cached by clients, gateways, or edge resolvers before requiring re-validation

    TTL in AgentAddr (Index Layer): The AgentAddr object returned by the index includes a ttl field. This defines how long the record may be cached by clients, gateways, or edge resolvers before requiring re-validation. It applies to the entire AgentAddr object, including: Metadata pointers (facts_url, private_facts_url) and optional adaptive_resolver_url. V0...

    work page

Show all 35 references
  1. [9]

    This TTL governs: How long the capabilities , telemetry endpoints , or evaluations may be considered fresh and When clients should re-fetch the facts for updated data

    TTL in AgentFacts (Metadata Layer): Each AgentFacts includes its own ttl value, often embedded in its cryptographically signed metadata section. This TTL governs: How long the capabilities , telemetry endpoints , or evaluations may be considered fresh and When clients should r...

  2. [10]

    Common TTLs in this layer could be - Static endpoints: 1,6 hours, Rotating endpoints: 5,15 minutes, and Adaptive routing tokens: 30,60 seconds

    TTL for Routing Metadata (Resolution Layer): Endpoint lists (static or rotating) and adaptive_resolver_url routing logic may have shorter TTLs than AgentFacts metadata. Common TTLs in this layer could be - Static endpoints: 1,6 hours, Rotating endpoints: 5,15 minutes, and Adap...

  3. [11]

    Rotating Endpoints: A set of URLs with short TTLs, designed for infrastructure that undergoes frequent restarts, redeployments, or geographical rebalancing

  4. [12]

    endpoints

    Adaptive Resolver URI: Points to a microservice or gateway that dynamically routes traffic to optimal downstream instances. Functions as a programmable interface for applying routing logic based on input features (location, time, capability match). AgentFacts may define one or...

  5. [13]

    ○ AgentFacts are hosted within their infrastructure (PrimaryFactsURL)

    Enterprise-Controlled Registries ○ Enterprises deploy and manage internal registries governing their agent namespaces. ○ AgentFacts are hosted within their infrastructure (PrimaryFactsURL). ○ Credential authorities reside within the enterprise’s compliance and audit teams

  6. [14]

    ○ AgentFacts reference third-party capabilities and certifications from accredited bodies

    Federated Industry Registries ○ Vertically integrated industries (e.g., logistics, healthcare) operate consortium registries. ○ AgentFacts reference third-party capabilities and certifications from accredited bodies. ○ Governance rules are encoded in index policy and enforced ...

  7. [15]

    ○ Credential validation and revocation are enforced via smart contracts or decentralized protocols

    Decentralized Public Registries ○ Publicly accessible registries allow open registration of agents. ○ Credential validation and revocation are enforced via smart contracts or decentralized protocols. ○ AgentFacts are hosted in IPFS or similar decentralized file systems, emphas...

  8. [16]

    V0.3, Work in Progress, Request for Comments

    Hybrid Topologies ○ Organizations may register with both private and public registries for broader discoverability. V0.3, Work in Progress, Request for Comments. Draft ○ Trust rules and resolution pathways allow clients to prioritize index sources based on use-case-specific lo...

  9. [17]

    This includes modeling hierarchical and mesh-based governance for industry-specific applications

    Governance Frameworks and Revocation Models We will formalize revocation protocols, dispute resolution mechanisms, and multi-jurisdictional governance workflows to enforce trust at scale. This includes modeling hierarchical and mesh-based governance for industry-specific applications

  10. [18]

    Open Standards and Interoperability Standardizing the AgentFacts schema with IETF/W3C bodies and enabling interoperability with Open Agent Frameworks, and existing Web3 identity platforms will ensure ecosystem adoption

  11. [19]

    Decentralized Identifiers (DIDs) v1.0,

    Privacy Enhancements with Zero-Knowledge Proofs (ZKPs) To further privacy guarantees, ZKP-based credential assertions [17] can be introduced, allowing agents to prove capabilities or compliance without revealing full certificate chains. XI. OPEN QUESTIONS AND DISCUSSION While ...

    2022

  12. [20]

    A Formal Treatment of Onion Routing,

    J. Camenisch and A. Lysyanskaya, "A Formal Treatment of Onion Routing," in Advances in Cryptology , CRYPTO 2005 , Lecture Notes in Computer Science, vol. 3621, Springer, 2005, pp. 169-187

    2005

  13. [21]

    IPFS - Content Addressed, Versioned, P2P File System,

    J. Benet, "IPFS - Content Addressed, Versioned, P2P File System," arXiv:1407.3561 , Jul. 2014. [Online]. Available: https://arxiv.org/abs/1407.3561

    2014 arXiv

  14. [22]

    The Stellar Consensus Protocol: A Federated Model for Internet-Level Consensus,

    D. Mazières, "The Stellar Consensus Protocol: A Federated Model for Internet-Level Consensus," Stellar Development Foundation, Nov. 2015

    2015

  15. [23]

    The Inevitable Rise of Self-Sovereign Identity,

    A. Tobin and D. Reed, "The Inevitable Rise of Self-Sovereign Identity," Sovrin Foundation, 2016

    2016

  16. [24]

    Chord: A Scalable Peer-to-peer Lookup Protocol for Internet Applications,

    I. Stoica, R. Morris, D. Karger, M. F. Kaashoek, and H. Balakrishnan, "Chord: A Scalable Peer-to-peer Lookup Protocol for Internet Applications," in Proc. ACM SIGCOMM , San Diego, CA, USA, Aug. 2001, pp. 149-160

    2001

  17. [25]

    Project NANDA. (2025). Upgrade or switch: The need for new index architecture for the internet of AI agents . https://tinyurl.com/nandappr1

    2025

  18. [26]

    (2025, May 20)

    Google. (2025, May 20). What's new with Agents: ADK, Agent Engine, and A2A Enhancements . Google Developers Blog. https://developers.googleblog.com/en/agents-adk-agent-engine-a2a-enhancements-google-io/

    2025

  19. [27]

    Anthropic. (2024). Introducing Model Context Protocol . Anthropic. https://www.anthropic.com/news/model-context-protocol

    2024

  20. [28]

    (2025, May 19)

    Microsoft. (2025, May 19). Introducing NLWeb: Bringing conversational interfaces directly to the web . Microsoft News. https://news.microsoft.com/source/features/company-news/introducing-nlweb-bringing-conversational-interfaces-directly-to-the-web/

    2025

  21. [29]

    Liu, J., Yu, K., Chen, K., Li, K., Qian, Y., Guo, X., Song, H., & Li, Y. (2025). ACPs: Agent Collaboration Protocols for the Internet of Agents . arXiv. https://arxiv.org/abs/2505.13523

    2025 arXiv

  22. [30]

    (2025, May 28)

    IBM Research. (2025, May 28). An open-source protocol for AI agents to interact . https://research.ibm.com/blog/agent-communication-protocol-ai

    2025

  23. [31]

    Agent Network Protocol Team. (2025). Agent Network Protocol Framework: A Protocol Framework for Agent Communication . https://agentnetworkprotocol.com/en/specs/01-agentnetworkprotocol-technical-white-paper/

    2025

  24. [32]

    S., Habler, I., & Sheriff, A

    Huang, K., Narajala, V. S., Habler, I., & Sheriff, A. (2025). Agent Name Service (ANS): A Universal Directory for Secure AI Agent Discovery and Interoperability . arXiv. https://arxiv.org/abs/2505.10609

    2025 arXiv

  25. [33]

    (2025, March 6)

    David, E. (2025, March 6). A standard, open framework for building AI agents is coming from… VentureBeat. https://venturebeat.com/ai/a-standard-open-framework-for-building-ai-agents-is-coming-from-cisco-langchain-and-galileo/

    2025

  26. [34]

    Ranjan, R., Gupta, S., & Singh, S. N. (2025). LOKA Protocol: A decentralized framework for trustworthy and ethical AI agent ecosystems . arXiv. https://arxiv.org/abs/2504.10915

    2025 arXiv

  27. [35]

    id": "nanda:550e8400-e29b-41d4-a716-4466554400

    Goldwasser, S., Micali, S., & Rackoff, C. (1985). The knowledge complexity of interactive proof systems. SIAM Journal on Computing, 18 (1), 186–208. https://doi.org/10.1137/0218012 V0.3, Work in Progress, Request for Comments. Draft APPENDIX Full AgentFacts Schema Fields marke...

    1985 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.