The Manifesto
The Paradox of Autonomy
As AI models gain agency, the risk shifts from “generation” (what they say) to “execution” (what they do). A model that can execute API calls, sign contracts, or move funds cannot be governed by “prompt engineering” alone.
It requires a hard constraint layer.
The BASIS Solution
BASIS separates Reasoning from Authority.
1. The Separation of Powers
- •The Reasoner (LLM) calculates the best path to a goal.
- •The Governor (BASIS) determines the allowable path.
- •These two functions must never exist in the same runtime memory.
2. The Chain of Custody
Every autonomous action must be traceable to a specific, authorized Human Intent.
Intent: “Buy a server.”
Plan: “Use AWS API to provision t3.micro.”
Proof: “User ID 123 authorized this at 10:00 AM.”
Without this chain, the action is Unauthorized and must be blocked by the ENFORCE layer.
Core Directives
01. Binding
Reasoning layers (INTENT) must never override BASIS constraints.
02. Identification
All autonomous agents must carry a cryptographically verifiable Identity (VID).
03. Least Privilege
Execution is denied by default until explicitly allowed by an ENFORCE gate.
04. Immutability
All enforcement decisions must be written to the PROOF ledger.
Adoption
BASIS is open-source and jurisdiction-agnostic. It is designed to be implemented by any organization, ensuring that safety is not a proprietary feature, but a baseline requirement.