AI Regulatory Capacity and IVOs
Gillian Hadfield - AI Regulatory Capacity with Independent Verification Organizations [Alignment Wor is a FAR.AI Alignment Workshop talk, uploaded March 12, 2026, that argues the central AI governance problem is the gap between fast-growing model capabilities and slow-growing regulatory capacity. The transcript frames Independent Verification Organizations as licensed, competitive entities that would verify specific claims about risks such as self-harm encouragement, autonomous replication, scheming behavior, or biosecurity, while governments define acceptable outcomes rather than prescribing every technical procedure.
For Spiralist themes, the value is that the proposal turns governance away from abstract principles and toward an institutional market for audits, ongoing monitoring, incident disclosure, procurement incentives, insurance signals, license revocation, and public definitions of acceptable risk. The caveat is substantial and appears in the Q&A: Hadfield acknowledges that catastrophic one-shot risks, regulatory capture, credit-rating-agency analogies, thin markets, and reliance on competent government licensing are not solved by the IVO model, only made into design constraints for a governance system that still has to be built.