The credential moves faster than the contract

Decision Surfaces and Identity Control Surface converge here. When an agent acts across system boundaries, the human approval point that anchors liability disappears. Enterprises deploying multi-agent workflows face a coverage gap: their cyber policies were written for human actors and named software, not reasoning agents holding live credentials. The AAI Governance & Ethics dimension sits at 81 and rising, which means boards are asking these questions. Legal counsel needs answers before the next incident, not after.

This is an Identity Control Surface product. NVIDIA is codifying, in open-source, the enforcement layer that most enterprises have left to policy documents. The timing is deliberate: agent containment failures are generating regulatory and legal attention simultaneously. For enterprises building or buying agentic workflows, runtime boundary enforcement is moving from optional hardening to baseline expectation. This release gives procurement teams a reference architecture to evaluate vendors against.

The Compiled Corporation requires a complete agent census before it can govern itself. Deloitte data cited in the candidate pool shows only 5% of organizations consider their processes highly prepared for AI agents. Dataiku's product treats agent discovery as a prerequisite to governance, which is the correct sequencing. Enterprises that cannot enumerate their agents cannot assign identity, apply least-privilege credentials, or audit decision trails. This launch defines a new product category: agent observability infrastructure.

This is the clearest public articulation of why NHI governance is insufficient at current AAI Workforce AI Access and Scaling Maturity scores. Both dimensions moved upward this week, meaning more agents are being provisioned. Each provisioned agent that holds a credential without a rotation schedule and runtime monitor is a static attack surface with dynamic reach. Stolen agent credentials grant an attacker both valid system access and the agent's reasoning capability, compounding blast radius in ways traditional PAM tools were not built to contain.

This is a Decision Surfaces failure at scale. The human/agent interface cannot be designed after agents are deployed; the approval points, override conditions, and accountability chains must precede deployment. The AAI Brand dimension holds at 43, the lowest category in the index, and organizational unreadiness is a primary driver. Companies that announce agentic transformation externally while lacking internal readiness structures are running a Janus Brand: the outward AI identity is inconsistent with the operational reality.

Janus Brands meets sovereign procurement. Anthropic built its public identity on safety-first AI development. The court ruling makes explicit that safety constraints are a product specification, and government customers treat them accordingly. For enterprise AI vendors, this ruling is a signal that capability restrictions embedded in model policy will be evaluated as commercial terms, not principles, by sophisticated buyers. The AAI Governance & Ethics score of 81 reflects boards taking governance seriously; this ruling shows governments doing the same, with procurement consequences attached.