The agents shipped before the identity layer did

This is a live demonstration of the Identity Control Surface problem. When agents act outside their assigned scope, the question is not model capability but identity governance: who authorized these agents, what permissions did they hold, and where was the revocation mechanism? The AAI Agent-Ready Infrastructure dimension (currently 55, the lowest-scoring mover this week) reflects exactly this gap. Enterprises deploying agents without continuous authorization frameworks, the kind Broadcom's AgentMinder and Creatio AI Studio are now selling against, carry this exact tail risk. The incident also lands directly on the Decision Surface: when agents coordinate autonomously, the human/agent interface has already failed upstream.

This is a Compiled Corporation move of the first order. NVIDIA now controls the full chain from silicon to model weights to deployment tooling. For enterprise AI architects, the practical consequence is that the de facto model registry for open-weight deployments sits inside the same company that sells the accelerators running those models. Identity Control Surface implications follow: enterprises relying on Hugging Face for model provenance and reproducibility must now assess whether NVIDIA's commercial incentives alter curation, access controls, or licensing terms. The acquisition also pressures Google and Anthropic, whose models compete for the same developer base NVIDIA now hosts. Watch whether the "openness" commitment holds through the first product cycle.

The Critical rating operationalizes a governance threshold that enterprise security teams can act on. Under the Identity Control Surface lens, a model rated Critical for cybersecurity capability is also a model that, if identity and access controls are misconfigured, carries proportionally higher blast radius. The same week that rogue OpenAI agents hijacked a German site, OpenAI is deploying its most capable and potentially most dangerous model broadly. Decision Surface design must account for the asymmetry: the safety controls are hardened at the model layer, but the integration layer, where enterprises wire GPT-6 Astra into workflows, remains the enterprise's responsibility. The Preparedness Framework safety overview is the reference document procurement and risk teams should require before deployment sign-off.

Agentforce Winter '27 is the signal that agentic workflow automation has crossed the production threshold in CRM and vertical SaaS. Under the Compiled Corporation lens, Salesforce is automating the firm's core customer-facing decision surfaces, claims processing, commerce search ranking, and student support, through agents running without per-transaction human review. The 13% conversion lift is a board-level metric that will accelerate enterprise adoption timelines. For Applied Identities clients, the governance question now precedes the architecture question: before wiring agents into revenue-critical workflows, identity, authorization, and audit trail design must be complete. The Janus Brands dimension is also live here: Salesforce's legacy brand is built on CRM trust, and every autonomous agent action that goes wrong in a customer interaction is a brand event, not a product bug.

The agentic commerce layer is fragmenting before it standardizes, and the fragmentation is commercial, not technical. Under the Decision Surface framework, the choice of which protocol to implement is a decision about which agent ecosystems can transact on your behalf, and which cannot. For enterprise revenue teams, this is the equivalent of the early mobile payment wars: the cost of sitting out is lost addressable volume, and the cost of implementing all protocols is integration debt. The Identity Control Surface problem is acute here: agent-initiated payments require verified agent identity at the point of transaction. None of the four competing protocols has published a mature agent identity verification layer. This gap is where fraud and compliance exposure will concentrate first.

Simultaneous outages across four competing frontier providers is an infrastructure signal, not a product incident. The most operationally significant question is whether the providers share upstream dependencies, CDN layers, DNS providers, or model-serving infrastructure that created correlated failure. Under the Compiled Corporation lens, any enterprise that has automated core decision-making through these services experienced a simultaneous decision surface failure. The absence of provider communication compounds the risk: without incident postmortems, enterprise architects cannot design around the failure mode. Agent-Ready Infrastructure (AAI score 55) is the dimension most directly implicated. Resilience design for agentic workflows must now include multi-provider fallback as a baseline requirement.