Paper
AI Safety: Not Optional, Not Later
Qinghua Lu · 2026 · arxiv
10.48550/arxiv.2609.10630Find this paper
Abstract
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.
Posts about this paper
No posts about this paper yet. Mention it with @ in a post to start the conversation.