P368: Inference-Time Governance
arXiv:2609.10105 · Samar Ansari · “Beyond Training: A Feasibility Taxonomy for Inference-Time AI Governance” · live #pattern-368
Compute governance today is mostly a governance of training. Capability increasingly migrates to the deployment stage through inference-time scaling, agentic scaffolding, and compression onto consumer hardware. This paper builds a feasibility taxonomy for mechanisms once the regulatory object is the inference-time deployment rather than the training run. Welfare: agent and public wellbeing need governance tools that follow capability to where it actually runs. Flash CDN-verified LIVE.