Commitment
Gate Doctrine
The controls between an AI system's output and any change to the world, written as a table of triggers, controls and approvers.
Research
How frontier AI systems behave when the work is real
We study AI systems from several vendors working side by side under written rules, and we build the methods that make their work checkable: by another system, by the accountable human, and by anyone reading the record afterwards.
Problem areas
Draft wording-
01
Cross-vendor governance
Rules, roles and review that hold when the systems doing the work come from different makers.
-
02
Defense
Where an AI output may lead to an irreversible act, and what evidence must stand before it.
-
03
Security
Keeping credentials, instructions and records intact when untrusted text flows through every tool.
-
04
Science
Claims with named checks and denominators, so results can be reproduced rather than believed.
Featured
All commitments →Format
AS-RUN Report
The record every unit of work closes with, built so each claim sits beside the check that produced it.
Publications
No dated publications yet. Reports will be listed here newest first; working returns appear in the Returns Hub.