HSP: The Human Supervision Protocol
Just open-sourced HSP Protocol - a fail-closed execution layer that enforces human oversight on autonomous AI.
The Problem
Current AI safety (RLHF, Constitutional AI) is 99.9% effective. But 99.9% means 1 in 1,000 actions can fail. For autonomous AI in finance/healthcare, that's unacceptable.
The Solution
HSP flips the model:
- Default state: BLOCKED (not allowed)
- Actions only execute after human cryptographic signature
- Every approval recorded on blockchain
Features
- Smart contract on Polygon (live)
- JavaScript & Python SDKs
- Mathematical risk scoring framework
- EU AI Act compliant
Links
- GitHub: github.com/jaquelinejaque/hsp-protocol
- Contract:
0x1BCe4baE2E9e192EE906742a939FaFaec50A1B4e - Patent: PCT/US26/11908
Free for personal/academic/OSS use!
Top comments (0)