SOURCE-LINKED INTELLIGENCE
AI Safety: Not Optional, Not Later
Incidents show that AI safety failures often arise across multiple layers. We present a safety-by-design assurance architecture combining model-level supervision, such as Scientist AI, with system-level controls over scaffolds and harnesses, independent verification, monitoring, and evidence infrastructure, supported by governance for accountability and evidence interoperability.
Read original source ↗ Open in workspace
- recordType
- paper
- region
- Global
Evidence & attribution
- arXiv · AI, language, vision and robotics · 2026-09-09T06:36:05.000Z
First collected: 2026-09-20T19:52:05.078Z. This is not the publication date.