The identity layer arrived before the trust problem was solved

This is a Decision Surfaces and Identity Control Surface move. Access management at the perimeter is insufficient when agents take multi-step actions inside live systems. Inline authorization means the governance boundary travels with the agent, which is the architectural requirement for any production agentic deployment. Enterprises still issuing static credentials to agents should treat this as a forcing function.

Short-lived credentials tied to verified agent identities address the core Identity Control Surface gap in current enterprise deployments, where agents either inherit overly broad human credentials or run without identity at all. The HashiCorp Vault integration means IBM is plugging into an existing secrets-management workflow rather than requiring a net-new toolchain. Paired with the Okta signal above, the market is converging on a standard pattern: verified identity plus scoped, ephemeral authorization.

This is the most operationally significant research finding in this cycle. The Decision Surfaces implication is direct: when agents are positioned as analytical aids, they become de facto decision-makers for the majority of users. Enterprises deploying agents for finance, legal, or strategy analysis need to treat agent output framing as a governance variable, not a UX detail. The 86% figure should be in every AI deployment review that involves subjective judgment.

This is a Janus Brands and Compiled Corporation signal. Oracle enterprise workflows entering the ChatGPT interface means the conversational layer becomes the primary decision surface for ERP-class processes. For Oracle customers, the brand experience of their core business systems now runs through OpenAI's interface, a channel they do not control. Organizations need to assess which workflows are exposed through this integration and where human approval gates remain.

SAP shipping agent suites inside core financial workflows is a Compiled Corporation signal. Treasury and accounts receivable are not peripheral processes; they are capital allocation functions. The multi-agent architecture for Cash and Treasury, three agents operating in coordination, raises Identity Control Surface questions: each agent in that suite requires its own identity, authorization scope, and audit trail. SAP customers should be asking for that architecture documentation before production rollout.

The NOA failure is a Janus Brands case study. A public celebration of an internal tool that did not work erodes staff credibility in AI initiatives and signals to talent that leadership is performing adoption rather than managing it. The author-publisher friction layer adds a second dimension: AI deployment decisions inside large publishers are now affecting external creative relationships. Both patterns, overstated internal launches and unmanaged external effects, are predictable failure modes for organizations without an identity architecture for their AI programs.