Field Insights

AI safety evaluations must account for cascading effects across the algorithmic supply chain

Current AI safety audits evaluate components in isolation, meaning the class of harms most likely to produce liability — those that emerge from how components interact across a supply chain — are systematically invisible to the frameworks regulators and courts will use to assign accountability.

OpenAI framework maps 18% of US jobs at higher near-term automation risk, 24% facing task displacement

OpenAI's own labor economist has now put a number on near-term automation risk across the US workforce — 18% of jobs at higher short-term risk, 24% facing task displacement — and that framework will show up in WARN Act analyses, workforce reduction litigation, and employer AI deployment defenses before most legal teams have read it.

Societal AI research has become less interdisciplinary, narrowing governance knowledge

The research base informing AI regulation has narrowed as the field grew — meaning the evidence behind compliance requirements may reflect disciplinary blind spots more than the actual risk landscape.