Admissibility & Evidence

Admissibility & Evidence

RESEARCH PROGRAMME — Governance. The Institute’s core methods programme. Grounded in the present and in dated, citable evidence.

Admissibility & Evidence

This programme studies the decision that comes before deployment: whether an AI capability may enter a consequential decision chain at all, and what record it must carry to earn that entry. It develops admissibility as a working method — a pre-runtime, record-based, reversible decision with four outcomes (Admit, Admit-with-Limits, Hold, Refuse) — and the evidentiary standard beneath it: no record, no standing.

Most AI governance begins after arrival — it regulates the machine, audits it, or responds to harm. This programme begins earlier, at the threshold, and asks what evidence a system must produce before it is allowed near a real decision. The “Evidence” half of the name is not decorative: admissibility is only as strong as the record it rests on. A capability that cannot be reconstructed — what it is, where it can act, who is accountable, how it can be stopped — has not earned standing, whatever its performance. The programme’s practical work is turning that principle into tests a regulator, auditor, or deployer can actually apply.

How it connects. Admissibility is the method the Institute proposes for the condition it calls synthocracy — the migration of decision-making power into and through AI systems. Where synthocracy names the problem, this programme builds the gate. See What Is Admissibility? for the concept, and What Is Synthocracy? for the condition.

Work in this programme

  • Admissibility: The Decision That Comes Before Safety — the foundational working paper: admissibility as a pre-runtime access decision, its four outcomes, and the no record, no standing rule.
  • Admissibility for AI Agents: A Record-Based Test — extends the method to autonomous agents through a record-based test (identity, authority, reach, delegation, human boundary, reversibility).
  • Capability Is Not Authority — why the power to do something is not the standing to do it: the distinction the whole programme turns on.
  • Colorado’s SB 189: A Law Rewritten to Disclose Rather Than Decide — a live case read through admissibility: a state that removed the pre-entry gate and kept only disclosure.
  • The Eighteen Days and The Switch Is the Story — the Fable/Mythos episode as the first public admissibility crisis: refusal that arrived too late, without standing on the record.

What we are working toward. A democratic pre-runtime gate that can handle legitimate secrecy without becoming secrecy; re-admission standards (no new evidence, no new status); and a minimum record that makes a decision contestable without demanding disclosure that increases risk.



Synthocracy Institute — Power & Accountability When AI Co-Decides