Skip to content

Human Supervision Protocol - Fail-Closed Execution Layer for Autonomous AI. Patent PCT/US26/11908

License

Notifications You must be signed in to change notification settings

jaquelinejaque/hsp-protocol

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

HSP: Human Supervision Protocol

The Fail-Closed Execution Layer for Autonomous AI

License Patent Mainnet EU AI Act


The Problem

AI Agents are moving from "chatbots" to "autonomous workflows" in Finance, Legal, and Healthcare. But there's a critical blocker: Liability.

  • A 99.9% success rate means 1 in 1,000 transactions is a lawsuit
  • Enterprises cannot deploy agents that rely solely on probabilistic safety filters (RLHF)
  • The EU AI Act (Regulation 2024/1689) mandates human oversight for high-risk AI systems

Current solutions are soft. HSP is hard.


The Solution: Proof of Supervision

HSP is not a filter. It is an Interceptor.

┌─────────────┐     ┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   AI Agent  │────▶│  HSP Layer  │────▶│   Human     │────▶│  Execution  │
│  (Request)  │     │ (Intercept) │     │ (Approval)  │     │  (Action)   │
└─────────────┘     └─────────────┘     └─────────────┘     └─────────────┘
                           │
                           ▼
                    ┌─────────────┐
                    │  Blocked    │
                    │ (No Bypass) │
                    └─────────────┘

Key Properties:

  • Fail-Closed: Actions are blocked by default until explicitly approved
  • Non-Bypassable: The AI cannot circumvent the supervision layer
  • Cryptographic Proof: Every approval is signed and immutable on-chain

Quick Start

Installation

# JavaScript/TypeScript
npm install @hsp-protocol/sdk

# Python
pip install hsp-protocol

Basic Usage

import { HSPClient } from '@hsp-protocol/sdk';

const hsp = new HSPClient({
  contractAddress: '0x1BCe4baE2E9e192EE906742a939FaFaec50A1B4e',
  network: 'polygon'
});

// Request supervised action
const action = await hsp.requestAction({
  type: 'FINANCIAL_TRANSFER',
  amount: 50000,
  recipient: '0x...',
  justification: 'Quarterly vendor payment'
});

// Action is pending until human approves
console.log(action.status); // 'PENDING_SUPERVISION'

// After human approval (signed cryptographically)
console.log(action.status); // 'APPROVED'
console.log(action.proofHash); // '0x7f3a...'

Mathematical Framework

HSP uses a rigorous mathematical model to calculate supervision requirements:

Risk Score (ρ)

ρ = α·A·I·(1-R) + β·(1-C)·D + (1-α-β)·T·F

Where:

  • A = Autonomy Level [0,1]
  • I = Impact Severity [0,1]
  • R = Reversibility [0,1]
  • C = Model Confidence [0,1]
  • D = Data Sensitivity [0,1]
  • T = Time Criticality [0,1]
  • F = Frequency (normalized)
  • α = 0.618 (Golden Ratio inverse)
  • β = 0.382 (φ⁻²)

Supervision Intensity (σ)

σ = 1 / (1 + e^(-γ·(ρ - 0.5)))

Where γ = 4.669 (Feigenbaum constant)

Full Mathematical Framework →


EU AI Act Compliance

HSP maps directly to EU AI Act requirements:

EU AI Act Article HSP Implementation
Article 14 (Human Oversight) Fail-closed approval mechanism
Article 12 (Record-Keeping) Immutable blockchain audit trail
Article 9 (Risk Management) Mathematical risk scoring
Annex IV (Documentation) Automated compliance reports

Compliance Documentation →


Architecture

┌────────────────────────────────────────────────────────────┐
│                      APPLICATION LAYER                      │
│            (Your AI Agent / LLM / Autonomous System)        │
└─────────────────────────────┬──────────────────────────────┘
                              │
                              ▼
┌────────────────────────────────────────────────────────────┐
│                       HSP SDK LAYER                         │
│    ┌──────────────┐  ┌──────────────┐  ┌──────────────┐    │
│    │   Observer   │  │   Validator  │  │  Supervisor  │    │
│    │    Layer     │  │    Layer     │  │    Layer     │    │
│    └──────────────┘  └──────────────┘  └──────────────┘    │
└─────────────────────────────┬──────────────────────────────┘
                              │
                              ▼
┌────────────────────────────────────────────────────────────┐
│                   BLOCKCHAIN LAYER (Polygon)                │
│              Contract: 0x1BCe4baE2E9e192EE906742a939FaFaec  │
└────────────────────────────────────────────────────────────┘

Deployed Contracts

Network Address Status
Polygon Mainnet 0x1BCe4baE2E9e192EE906742a939FaFaec50A1B4e Live
Polygon Amoy (Testnet) Coming Soon -
Ethereum Mainnet Coming Soon -

Documentation


Use Cases

Financial Services

Autonomous trading agents with mandatory human approval for transactions above threshold.

Healthcare AI

Diagnostic AI systems with physician sign-off requirements.

Legal Tech

Contract analysis AI with lawyer supervision for binding decisions.

Enterprise Automation

RPA bots with escalation paths for edge cases.


Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.


License & Patents

This project is licensed under the Apache License 2.0 - see LICENSE.

Important: The HSP Protocol is covered by international patent application PCT/US26/11908. Commercial use requires a license. See PATENTS.md for details.


Contact


Building the trust layer for autonomous AI.

About

Human Supervision Protocol - Fail-Closed Execution Layer for Autonomous AI. Patent PCT/US26/11908

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •