Risk LabReasoned from system designv1.15.0
In plain English
This page provides local browser worksheets. They help plan reviews; they are not formal safety certifications.
- Why this matters: AI risk can come from the whole arrangement, not one obvious model.
- What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
- Technical version below: the expert terminology remains available and is linked through the glossary.
Most Likely Threat Review Worksheet
Evidence levelReasoned from system designTechnical label: Architectural inference
This worksheet helps reviewers identify whether an AI system has the conditions for distributed behavioral persistence. It is not a certification method and does not calculate real-world probability.
Most Likely Threat Review
Planning worksheet only. It checks conditions for distributed behavior persistence; it does not calculate probability or certify safety.
How to use the result
If the worksheet reports elevated concern, do not treat it as proof of harm. Treat it as a prompt to gather missing evidence:
- exact composition manifestA machine-readable record of the exact runtime composition used for an evaluation, release, incident, or rollback. Open glossary definition;
- route-specific evaluation;
- memory and synthetic-data provenanceA record of where a component or behavior came from. Open glossary definition;
- evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition independence record;
- descendant and alias history;
- rollbackReturning a system to an earlier known state. Open glossary definition packet;
- responsibility map;
- behavioral-extinction review.
The purpose is to identify where ordinary component review is insufficient.