AIIC AI Intelligence Centre

SOURCE-LINKED INTELLIGENCE

AI Safety: Not Optional, Not Later

arXiv · AI, language, vision and robotics · article · Sep 9, 2026 · UTC

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

First collected: 2026-09-20T19:52:05.078Z. This is not the publication date.