AI & ML · Oct 6
OpenAI, Meta, Anthropic and Google could not quantify worst-case AI risk
At a New York City Council hearing, frontier labs exposed the distance between deploying powerful models and measuring their most severe risks.
- Do not accept frontier AI safety claims without asking what is measured, audited, and monitored after deployment.
- Builders should connect evals, incidents, and deployment gates into an auditable risk record.
- Policymakers should demand measurable safety evidence without reducing AI governance to paperwork theater.