The Agents Shipped. The Identity Layer Didn't.

This is infrastructure for the Decision Surface layer — the exact seam where human oversight hands off to autonomous agent execution. Remote MCP support means agents can reach across organizational boundaries to external services, which immediately raises Identity Control Surface questions: which agent, running under whose credential, with what scope? Enterprises evaluating Google Cloud for agentic workloads need governance answers before they enable these capabilities, not after. The background task model also accelerates the Compiled Corporation thesis — decisions that once required human scheduling now execute on event triggers.

Nemotron Labs is a direct Janus Brand signal — NVIDIA is simultaneously a chip company and now an AI model steward, and those identities create visible tension in how enterprises perceive vendor lock-in risk. More importantly, the 'trust, control, customize' positioning maps directly onto Identity Architecture: enterprises that fine-tune open models on proprietary data have a materially different identity governance posture than API consumers. The open model path enables internal non-human identity registries tied to specific model versions — a meaningful control surface advantage. Watch whether Nemotron's brand promise survives contact with NVIDIA's closed hardware ecosystem.

Oracle's GA move closes a meaningful gap for enterprises that have resisted cloud-native AI stacks due to compliance and data residency concerns. The OpenAI API compatibility layer is a Decision Surface play — it lowers switching costs enough that procurement conversations shift from 'can we use AI' to 'which governed environment runs it.' The bundled agent development tooling signals that Oracle is building a Compiled Corporation substrate for its existing enterprise install base. The Identity Control Surface implication is significant: OCI's tenancy model means agent credentials can inherit existing IAM governance, which no hyperscaler-native offering matches cleanly today.

Prompt injection is the primary Identity Control Surface attack vector for agentic systems — it is how adversaries hijack agent identity and redirect tool calls. GPT-Red represents OpenAI operationalizing a defense at the model layer, but the enterprise implication is that application-layer defenses remain the customer's responsibility. No model-layer hardening eliminates the need for agent identity governance, scoped credentials, and audit trails in the deployment environment. Enterprises treating model safety improvements as a substitute for their own control surface work are exposed.

This is the Decision Surface becoming a marketplace. When agents from four major enterprise SaaS vendors can be dropped into a shared orchestration layer, the question of who governs non-human identities across that fleet becomes urgent and unanswered. Each vendor's agent arrives with its own credential model, scope assumptions, and audit behavior. Google's Agent Gallery simplifies deployment but does not solve cross-vendor identity federation — that gap lands squarely on enterprise architects. The Compiled Corporation upside is real: composable agent workflows across ERP, CRM, ITSM, and content systems are the operating model of the next enterprise generation. The governance debt is equally real.

For enterprise AI programs, a patchwork of state AI laws is an Identity Architecture compliance problem before it is a policy one. Non-human agents operating across state lines — executing transactions, making recommendations, processing personal data — will encounter inconsistent disclosure, accountability, and explainability requirements. OpenAI's advocacy signals that no unified federal standard is imminent, and enterprises building agentic systems today should assume multi-jurisdictional compliance obligations as a permanent condition. The Janus Brand dimension: OpenAI is simultaneously a model provider and a policy actor shaping the regulatory environment its customers must navigate — a dual identity that warrants scrutiny.