Skip to content

at1c-protocol/at1c-protocol-official

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

123 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AT1C Protocol

Cryptographic Human Consent for AI Agent Actions

MIT License npm Registry

AT1C gives you a cryptographically signed, independently verifiable proof that a human approved a specific AI agent action — before that action executes.

No approval, no valid action. No exceptions.

Built for developers. Required for EU AI Act compliance (enforcement: August 2026).


Try it in 30 seconds

git clone https://github.com/at1c-protocol/at1c-protocol-official.git
cd at1c-protocol-official
npm install
npm run demo

Or install the SDK directly:

npm install @at1c/sdk

Key Documents

  • 📄 Whitepaper — EU AI Act compliance, sector analysis, Algorand x402 payment architecture
  • 📜 Manifesto — AT1C founding principles: sovereign identity, humans first
  • 📚 All docs — protocol spec, receipts, verification, SDK reference

Live Registry

curl https://registry.at1c.com/health

How it works

  1. Request — an AI agent asks permission to perform a specific action
  2. Approve — a human explicitly grants or denies it
  3. Proof — an Ed25519-signed receipt is generated, binding the user, action, timestamp, and nonce
  4. Verify — any system independently verifies the receipt before execution — no trust required

Every receipt is single-use (nonce-based replay protection). Approval is scoped to the exact action requested — not transferable to any other action.


AT1C vs. The Others

The agentic AI space has several overlapping projects. Here is where AT1C differs from the closest ones:

vs. autonomous agent runtimes (Theseus, etc.) — those systems remove humans from the loop by design. AT1C keeps a human in the loop by design. These are not competing approaches — they target different buyers. Autonomous runtimes suit DeFi/crypto use cases where fully autonomous execution is the goal. AT1C suits regulated sectors where human oversight is legally required.

vs. data redaction gateways (TrustLayer, etc.) — those systems control what data AI sees. AT1C controls what actions AI is permitted to take. A developer could use both without conflict.

vs. payment authorization protocols (ATXP, etc.) — those systems authorize agent payments. AT1C authorizes any agent action, with cryptographic proof of human approval, across any domain — not scoped to payments.

The field is converging on a shared framing: treat autonomous agents as instruments acting under a person's authority, with scoped, revocable, auditable chains of responsibility. AT1C implements this today, with a working SDK and live registry.


Why it matters right now

The EU AI Act begins enforcement in August 2026. It requires:

  • Accountability for automated decisions
  • Human oversight of high-risk AI actions
  • Auditable records of AI behaviour

AT1C gives you all three out of the box. Primary target sectors: fintech, health, legal.


Core Safety Rules

  • No implicit authority — nothing acts without explicit approval
  • Context binding — approval is valid only for its exact action and resource
  • Replay protection — every receipt is single-use
  • Verification before execution — actions must be verified before they run
  • Non-custodial — AT1C never holds signing keys. Agent keys and user keys stay on their respective devices.

Project structure


Documentation


Known Limitations & Roadmap

  • Agent key custody — ✅ done (v1.1) — non-custodial by design. Agents generate their own keypairs locally; the registry only ever receives and signs over the public key.

  • Live registry — ✅ done (v1.2) — authenticated agent verification live at registry.at1c.com over HTTPS. Registry private key held in environment variables only, never on the filesystem.

  • End-user onboarding (planned) — passkey-based (WebAuthn/FIDO2) flow. The signing key lives on the user's device, unlocked by biometric — AT1C never holds it.

  • Receipt storage (planned, opt-in) — currently receipts are stored locally. Hosted long-term storage planned as an optional paid tier, framed as evidence/insurance retention, not a protocol requirement.

  • Tier validation (planned) — agent tier is currently caller-supplied with no server-side validation. Will be constrained to server-side assignment before open registration launches.

  • Quantum resistance (future) — AT1C currently uses Ed25519, which is not post-quantum secure. Migration to a NIST-approved post-quantum signature scheme is on the long-term roadmap. Not urgent for August 2026 compliance, but noted transparently.

  • Agent Manifest (future) — structured, signed per-agent document mapping onto EU AI Act Article 13/14 documentation requirements.

These are documented deliberately — AT1C's value depends on being trustworthy, and that includes being transparent about what is solid today versus what is still being built.


Licence

MIT — AT1C Protocol Contributors

About

AI Agents are verified using Human consent, produce receipts with cryptography based evidence, low cost insurance policy.

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages