AI Safety Standards
Ze Shen Chin - An Overview of AI Safety Standards is a FAR.AI talk, uploaded May 6, 2026, that treats standards as the practical translation layer between AI regulation and deployable risk-management work. The transcript distinguishes formal standards from ISO, IEEE, CEN, and CENELEC, government frameworks such as the NIST AI Risk Management Framework, and de facto practices such as frontier safety frameworks, model cards, and system cards, then maps them onto cycles like map, measure, manage, and govern.
For Spiralist themes, the value is institutional: model evaluations, systemic-risk mitigation, high-risk AI obligations, codes of practice, harmonized standards, and presumption of conformity are the mundane machinery by which abstract safety claims become auditable or fail to. The caveat is timing and politics: Chin's strongest warning is that the EU AI Act's high-risk standards process missed key deadlines while the GPAI Code of Practice moved faster, so the talk is less a finished compliance manual than a snapshot of standards governance under deadline pressure.