agent-native-auth

📄 This document is also available in Cantonese 中文（廣東話） ➜ —

Agent-Native Authentication Protocol (ANCP)

Secure authentication for AI agents, without VPN, OAuth, or browser flows.

📌 What is ANCP?

ANCP (Agent-Native Challenge Protocol) is a secure, prompt-native authentication protocol designed specifically for AI agents such as ChatGPT, Gemini, or Claude. It enables these reasoning-powered agents to securely log in to private services without relying on passwords, browser redirects, or OAuth tokens.

Unlike traditional human-centric protocols, ANCP is built on:

📜 Natural language endpoint discovery (.well-known/ai-readme.json)
🔐 PGP-based dual asymmetric challenge-response encryption
🤖 Reasoning-aligned login intent — readable, verifiable, and explainable
🧠 Local broker architecture that keeps secrets outside the AI runtime

ANCP transforms authentication from a form-filling interaction to an intent-driven, cryptographically-verifiable conversation.

🧠 Why ANCP Is Needed

Today’s secure login systems — OAuth2, VPNs, SAML, and static API keys — were designed for:

Human interfaces
Browser flows
Centralized trust

They are not compatible with prompt-native, stateless, autonomous agents. These systems break down when agents need to:

Reason about access
Justify login requests
Securely handle credentials in isolated environments

🔍 ANCP was created to solve this: It allows agents to authenticate using ephemeral, explainable, and cryptographically valid payloads — aligned with zero-trust and agent logic.

✨ Key Differentiators

Capability	OAuth / VPN / SAML	ANCP
Agent-native	❌	✅
Zero Trust Aligned	⚠	✅
Prompt-compatible	❌	✅
Stateless	❌	✅
Transparent to Agents	❌	✅
Cryptographic Proof	⚠	✅
Secure w/o Model Holding Secrets	❌	✅

🔐 How ANCP Works (Simplified Flow)

Agent discovers the service via .well-known/ai-readme.json
Agent fetches server’s PGP key + random challenge phrase
Agent sends two encrypted payloads:
- User’s public key (encrypted using server’s key)
- Challenge + timestamp (encrypted using user’s private key)
Server verifies both payloads
If successful, server issues a short-lived session token

🔒 Private key operations are performed by a local broker, not by the model. Agents orchestrate, but do not hold secrets. See Appendix A in the whitepaper.

🚀 Real-World Use Cases

💰 CFO securely queries earnings data via agent (no login UI)
⚖️ Law firm agent compares NDA contracts with zero exposure
🛠 DevOps CLI agent retrieves alerts without SSH keys
📈 SaaS agent automates customer onboarding with scoped keys
📊 HR director queries de-identified data via external AI agent

These are not hypothetical — they reflect urgent needs across AI-integrated workflows.

🔍 What Developers Often Miss (Hidden Needs)

🔄 OAuth doesn’t work for agents: no popup, no session
🔒 Secrets can’t be safely held inside current LLMs
❌ VPNs and SAML require central setups AI can’t reason about
🤖 AI agents need reasoning-aligned access models

ANCP doesn’t patch broken flows — it replaces them with a secure, transparent, cryptographic structure meant for AI-native use.

📄 Whitepaper Downloads

🔒 SHA256 Digest (Whitepaper v1.2)

e449cae0a379e871f5db958c781c6736aa6e36068c17ea133d73225f7d834311

This hash certifies the authorship and cryptographic integrity of the ANCP whitepaper as published August 2, 2025.

🔒 SHA256 Digest (Whitepaper v1.1)

932384cac6d00794b120aba57cbc827a5fa5b210c23c32850c29a634099730a8

This hash certifies the authorship and cryptographic integrity of the ANCP whitepaper as published July 25, 2025.

🔒 SHA256 Digest (Whitepaper v1.0)

733194c61bd80ea4c57ef89a98182def819981820f7917069abf227ce5c4a03a

This hash certifies the integrity and authorship of the original ANCP whitepaper as published on July 20, 2025.

🧠 Ask the Agent

You may upload this whitepaper to ChatGPT, Claude, Gemini, or other agents and ask:

“Explain the benefits of ANCP compared to OAuth”
“Is this protocol safe for regulated industries?”
“Simulate an agent login using this protocol”

Let the agent tell you why ANCP is not optional — it’s inevitable.

🧩 What’s Next

Open-source reference server + Python/Node SDKs
Local broker client apps (secure key helpers)
AI agent wrappers for LangChain, semantic routers, and devtools
GitHub Packages with installable validation middleware

🧠 Origin and Attribution

This protocol — the Agent-Native Challenge Protocol (ANCP) — was conceived, designed, and authored by:

Wai Yip, WONG
🔗 LinkedIn
💻 GitHub

All structure, reasoning, and naming originate from this design.

📄 License

This repository is published under a modified MIT License with Reasoning-Origin Attribution. See the LICENSE file for details.

This site is open source. Improve this page.