.@Deloitte 2026: 21% of enterprises report mature agentic AI governance. That leaves roughly four in five running agents on some mix of policy documents, human spot checks, and hope. In any other domain that touches capital, a 21% control-maturity rate would be an audit finding, not a benchmark. The reason it is tolerated is that agent liability has not yet been priced. Nobody has been sued, fined, or delisted over an agent acting outside mandate at scale. Yet. The wiki incident is an early data point on what unpriced looks like. The California Attorney General's investigation into the related Hugging Face intrusion is an early data point on what pricing looks like. Origins makes governance a property of the substrate rather than a maturity score on a survey. Identity, authority, and execution are verifiable whether or not the organization has a governance program. The 79% are not behind on process. They are ahead of the infrastructure. That gap closes from the infrastructure side.