The Upstream Safety System is a deterministic governance boundary that sits between permission and action.
It does not predict outcomes.
It does not optimize behavior.
It does not act on its own.
It exists for a single structural purpose: to ensure that no high‑risk action proceeds without declared human authority.
Built and tested
The USS was built as a working system and tested end‑to‑end as a deterministic governance layer. It is not a conceptual framework or advisory model. It is an executable architecture designed to operate as a choke‑point between permission and action.
During its construction, a structural constraint surfaced: no system can legitimately act without declared authority. That constraint led directly to the discovery of the Three Governance Primitives.
What it does
The USS enforces a single rule:
No high‑risk action proceeds without declared authority.
For any client‑defined high‑risk action, the system requires:
- Declared purpose
- Declared authority source
- Declared constraints
If these are not present, the action is structurally blocked.
Core structural invariant
Legitimacy is treated as a binary structural property of a decision‑permitting system. A system may still function without these conditions, but it does so without structural legitimacy.
L = δ × (A ∧ D ∧ C ∧ I)
- δ (Governance layer): declared purpose, named human authority, and explicit constraints.
- ADCI (Value layer): Agency, Dignity, Continuity, and Interpretive Authority.
This page is an orientation layer. The formal proofs and canonical definitions are archived in the public scholarly record. See Formal Records for the Zenodo sequence.
What it is not
- Not a moderation system
- Not an AI alignment layer
- Not a predictive safety model
- Not an automated decision engine
The USS is a governance boundary, not a behavioral optimizer.
Origin
The system was first built during an MLAI hackathon sponsored by the Office of the Australian eSafety Commissioner. It was intended to make AI systems safer by introducing a deterministic governance choke‑point.
During that build, the structural rule behind the system became visible: legitimate action requires declared authority. From that constraint, the Three Governance Primitives were discovered.
Contact
3primitives.io engages with select collaborators and institutions.
USS inquiries are open; implementation details and operational tooling are provided privately.