hsp-protocol 0.2.0

Human Supervision Protocol - Fail-closed AI governance for EU AI Act compliance

HSP Protocol SDK

Fail-Closed AI Supervision for EU AI Act Compliance

Overview

HSP Protocol provides the only patented fail-closed architecture for AI agent supervision that ensures EU AI Act Article 14 compliance.

Unlike probabilistic safety filters (RLHF, guardrails) that fail open, HSP implements cryptographic pre-execution approval that fails closed — no approval, no execution.

Supported Providers

Provider	Package	Status
Google Gemini	hsp-protocol[gemini]	✅ Production
OpenAI GPT	hsp-protocol[openai]	✅ Production
Azure OpenAI	hsp-protocol[openai]	✅ Production
Anthropic Claude	hsp-protocol[anthropic]	✅ Production
AWS Bedrock	hsp-protocol[bedrock]	✅ Production

Installation

# Core (minimal)
pip install hsp-protocol

# With specific provider
pip install hsp-protocol[gemini]
pip install hsp-protocol[openai]
pip install hsp-protocol[anthropic]
pip install hsp-protocol[bedrock]

# All providers
pip install hsp-protocol[all]

Quick Start

from hsp_sdk import HSPClient

# Initialize with any provider
client = HSPClient(
    provider="gemini",           # or "openai", "azure", "anthropic", "bedrock"
    organization="Acme Corp",
    ai_system="CustomerBot",
    risk_level="high"            # EU AI Act classification
)

# All interactions are now supervised
response = client.chat("Process refund for order #12345")

# Generate compliance report (RAT)
rat = client.generate_rat(anchor_blockchain=True)
print(f"Report: {rat['report_url']}")

How It Works

┌─────────────────────────────────────────────────────────┐
│                    YOUR AI AGENT                        │
│                         │                               │
│                         ▼                               │
│              ┌─────────────────────┐                    │
│              │   HSP INTERCEPTOR   │ ◄── Patented       │
│              │   (Fail-Closed)     │                    │
│              └─────────────────────┘                    │
│                         │                               │
│         ┌───────────────┼───────────────┐               │
│         ▼               ▼               ▼               │
│   [Risk Score]   [Human Approval]  [Crypto Sign]        │
│         │               │               │               │
│         └───────────────┼───────────────┘               │
│                         ▼                               │
│              ┌─────────────────────┐                    │
│              │  IMMUTABLE LEDGER   │ ◄── Polygon        │
│              │  (Proof of Superv.) │                    │
│              └─────────────────────┘                    │
│                         │                               │
│                         ▼                               │
│                 [EXECUTE / BLOCK]                       │
└─────────────────────────────────────────────────────────┘

Risk-Based Approval

Risk Score	Action	Quorum
0-40	Auto-approve + log	0
41-70	Single human approval	1
71-90	Multi-sig approval	2
91-100	Critical review	3+

Provider-Specific Usage

Google Gemini / Vertex AI

from hsp_sdk import HSPClient

client = HSPClient(
    provider="gemini",
    organization="HealthPlus",
    ai_system="MedicalDiagnosis",
    api_key="your-gemini-key"  # or GOOGLE_API_KEY env var
)

OpenAI

client = HSPClient(
    provider="openai",
    organization="TradingCo",
    ai_system="AlphaTrader"
    # Uses OPENAI_API_KEY env var
)

Azure OpenAI

client = HSPClient(
    provider="azure",
    organization="Enterprise",
    ai_system="CopilotCustom",
    azure_endpoint="https://myresource.openai.azure.com/",
    deployment="gpt-4o"
)

Anthropic Claude

client = HSPClient(
    provider="anthropic",
    organization="LegalFirm",
    ai_system="ContractReview"
)

AWS Bedrock

client = HSPClient(
    provider="bedrock",
    organization="FinanceOrg",
    ai_system="FraudDetector",
    region="us-east-1"
)

Compliance Reports (RAT)

Generate a Report of Action & Traceability:

rat = client.generate_rat(
    anchor_blockchain=True,  # Polygon anchoring
    sign_ecdsa=True          # Cryptographic signature
)

print(rat)
# {
#   "rat_id": "rat_hsp_abc123",
#   "actions": [...],
#   "approvals": [...],
#   "blockchain": {
#     "network": "polygon",
#     "contract": "0x1BCe4baE...",
#     "tx_hash": "0x..."
#   },
#   "compliance": {
#     "eu_ai_act_art_14": "COMPLIANT",
#     "iso_42001": "COMPLIANT"
#   }
# }

EU AI Act Compliance

HSP Protocol directly implements:

• Article 14: Human oversight with pre-execution approval
• Article 29: Transparency via immutable audit logs
• Article 9: Risk management through dynamic risk scoring
• Annex IV: Technical documentation via RAT reports

Pricing

Tier	Transactions/Month	Price
Starter	10,000	€499/mo
Business	100,000	€2,499/mo
Enterprise	Unlimited	€9,999/mo

Contact Sales

Patent Information

HSP Protocol is protected by international patent:

• PCT/US26/11908 (Priority Date: 2024)
• USPTO Pending: 63/948,692
• National phase entries: EP, CN, JP (2026-2027)

Commercial use requires licensing. Learn more

Support

• Documentation: https://docs.hsp-protocol.com
• Email: [e-mail]
• Enterprise: [e-mail]

---

HSP Protocol — The only architecture that makes AI compliance defensible.

Patent PCT/US26/11908 | © 2024-2026 Jaqueline de Jesus