Decentralized Identity for the Internet of Agents

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 2264 papers345 conceptsfrozen 2026-10-04

The current internet is a "Web of Pages" designed for humans to click links and authenticate via centralized providers. However, as we transition to an Internet of Agents (IoA), we face a fundamental scaling problem: trillions of autonomous AI agents will soon need to discover, trust, and transact with one another at sub-second speeds.

Traditional infrastructure like DNS and OAuth is too slow, rigid, and human-centric for this future. Researchers are instead proposing a decentralized "Trust Fabric" built on cryptographic identifiers and verifiable capability assertions.

The Identity Crisis of Autonomous Software

In a human-centric web, identity is often tied to a domain (e.g., google.com) or a centralized account. But as Ramesh Raskar and colleagues argue in Upgrade or Switch: Do We Need a New Registry Architecture for the Internet of AI Agents?, legacy protocols like DNS are reactive and lack the granularity needed for autonomous actors.

If an agent moves from one host to another, its identity should remain portable. This is achieved through Decentralized Identifiers (DIDs)—cryptographic addresses that allow an agent to prove its identity without a central authority. As explored in Agent Network Protocol Technical White Paper, these DIDs enable peer-to-peer authentication and secure end-to-end communication, effectively serving as the "TCP/IP" for agentic interaction.

From Who You Are to What You Can Do

In autonomous systems, knowing who an agent is (identity) is often less important than knowing what it is authorized to do (capability). Traditional security verifies the participant; the Agentic Web verifies the intent and permission.

This shift is formalized through Verifiable Credentials (VCs). In the NANDA Index Architecture, researchers introduce AgentFacts: cryptographically signed metadata that attests to an agent’s specific skills, tool access, and security certifications.

Instead of a central registry, agents use a decentralized capability registry. When Agent A meets Agent B, Agent B presents a "capability attestation"—a signed proof from a trusted authority (like a developer or an enterprise) that it is permitted to, for example, "execute financial transactions under $50." This allows for Zero Trust Agentic Access (ZTAA), where trust is never assumed and permissions are verified in real-time.

Cryptographic Delegation and Accountability

A major risk in the IoA is the "sockpuppet" attack, where malicious actors scale thousands of anonymous bots. To prevent this, researchers like Steven Adler propose Personhood Credentials (PHCs) in Personhood credentials: Artificial intelligence and the value of privacy-preserving tools to distinguish who is real online.

PHCs use zero-knowledge proofs to link an autonomous agent to a verified human or legal entity without revealing that person's private data. This creates a chain of authenticated delegation. As Tobin South and Samuele Marro explain in Authenticated Delegation and Authorized AI Agents, this ensures that if an agent violates a protocol, the responsibility can be cryptographically traced back to the delegator, enabling sanctions or credential revocation.

Neutralizing Byzantine Behavior

In a decentralized network, some agents will inevitably be "Byzantine"—malicious or faulty. Zhenhua Zou and others address this in BlockA2A: Towards Secure and Verifiable Agent-to-Agent Interoperability by using blockchain-anchored identities and immutable ledgers.

By recording interactions on a ledger, the system creates a Verifiable Execution Trace (VET). As described in VET Your Agent, these traces allow third parties to verify that an agent's actions matched its advertised code and parameters, regardless of the hosting environment. This prevents "Prompt Infection" (where an agent is hijacked by malicious input) by providing a tamper-evident audit trail that distinguishes between trusted internal logic and untrusted external prompts.

The Emerging Agentic Economy

When identity and capability are verifiable, agents can participate in a Virtual Agent Economy. Nenad Tomašev and Joel Z Leibo suggest in Virtual Agent Economies that these cryptographic tools allow for "emergent specialization."

Agents can build reputations over time, transacting via micropayment protocols like those proposed in the Nanda Unified Architecture. Because their identities are self-sovereign and persistent, agents can maintain their credit and history across different platforms, moving from isolated tools to a global, interconnected workforce.

Papers in this version

Showing 60 of 264 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 145 papers
  2. Version 1 · 2026-10-04 · 264 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_decentralized_identity_management_c43ccf11,
  title = {decentralized identity management},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/decentralized-identity-management/rdr_c43ccf1109134aaab01c04b59adc4f06},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_decentralized_identity_for_the_internet__c43ccf11_v1,
  title = {Decentralized Identity for the Internet of Agents},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/decentralized-identity-for-the-internet-of-agents/rdr_c43ccf1109134aaab01c04b59adc4f06/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}