IBM and SAP Push Agent Identity and Governance Into the Enterprise Stack

Two of the largest enterprise software vendors spent the week of October 26 selling the same idea: AI agents need identities, owners and an off switch before they can run real business processes. IBM announced agent identity, discovery and containment tools at its TechXchange conference in Atlanta on October 27, and SAP used its TechEd event to add an enterprise AI gateway and a governance agent to its Joule platform. For CIOs and chief AI officers, the announcements signal that agent governance is moving out of slide decks and into products they can buy.

What did IBM announce for AI agents?

IBM's announcement centers on giving each agent a verifiable identity and limiting what it can reach. HashiCorp Vault 2.2 gives agents their own secured accounts with automatically renewed credentials. A new Agent Identity capability keeps an authoritative record of every agent, including its owner, lifecycle status and granted authority, and feeds that to Vault to enforce least-privilege access and revoke credentials in real time. Both are generally available.

IBM also extended watsonx Orchestrate beyond its own ecosystem. A Discover feature, in tech preview, finds agents, models and MCP servers built on other platforms, including Amazon Bedrock, Microsoft Foundry and Google Gemini. A Contain feature, also in tech preview, runs agents in isolated sandboxes and provides a kill switch for any agent. An Evaluate capability, now generally available, tests agents for quality, accuracy and compliance before deployment and tracks cost by agent, model and token.

IBM said Agent Identity and Vault integrate with Nvidia's OpenShell secure runtime to give each agent a protected runtime and short-lived credentials. The company cited Gartner estimates that the average Fortune 500 enterprise will use more than 150,000 agents by 2028, and that only 13% of organizations believe they have the right agent governance in place.

What is SAP changing in Joule?

SAP's update aims to turn Joule from an assistant that answers questions into a layer that runs workflows, CIO reported. Joule Studio will be integrated with Joule Work, so users can describe what they want and have the system coordinate agents, apps and data. Joule Work provides 40 assistants that can direct hundreds of agents beneath them.

On governance, SAP announced an Enterprise AI Gateway that consolidates authentication, policy enforcement, routing, monitoring, usage tracking and auditing, plus a Governance Agent designed to help companies comply with the EU AI Act. Agents can reach outside systems through the Model Context Protocol and the Agent2Agent protocol, and third-party agents can call Joule agents through the governed gateway. SAP plans general availability of the gateway in the first quarter of 2027, according to CIO, while Joule Work and the new Studio features roll out in phases.

SAP chief AI officer Jonathan von Rüden described the appeal for business users who build their own workflows: "I'm not queuing for eight months in the IT backlog."

Why does agent identity matter now?

The common thread is accountability. When an agent acts with a shared service account or a borrowed human login, security teams cannot tell who owns it, what it is allowed to do or how to stop it. Giving each agent its own identity and short-lived credentials makes those questions answerable, and it makes agents auditable in the same way employees are.

Speakers at AI Connect events have been pressing this point for months. At COLLIDE 2026, IBM's Ann Leach argued that enterprises need an agentic control plane to manage agent sprawl, and she set a test that applies directly to this week's launches. "If the control works only inside one ecosystem, it's not really a control plane," she said. IBM's cross-platform Discover feature and SAP's MCP and A2A support are both attempts to pass that test, though both are early.

NexusOne CEO Anu Jain made a related case at the same conference for one identity across all systems. "AI needs complete context and hard boundaries," he said.

What should enterprise leaders do with these announcements?

First, inventory before buying. Tools that discover agents across platforms are only useful if someone owns the resulting list and decides which agents should keep running. Second, check what is actually shipping. Several of IBM's most relevant features, including Discover and Contain, are in tech preview, and SAP's gateway is not due until 2027.

Third, avoid trading agent sprawl for governance sprawl. Many enterprises run IBM, SAP, Microsoft and Salesforce side by side, and each vendor now offers its own agent controls. As Xnode's Hemakumar Velugula argued at The AI Enterprise Conference 2026, the hard part is proving control over what agents reach across the whole stack, not inside one product.

The next thing to watch is whether these identity and gateway layers interoperate in practice, or whether CIOs end up reconciling a separate agent registry for every major vendor.

Frequently asked questions

What did IBM announce for AI agent identity?

On October 27, 2026, IBM announced Agent Identity and HashiCorp Vault 2.2, which give agents their own accounts, owners and auto-renewed credentials, plus watsonx Orchestrate features to discover, contain and evaluate agents.

What is SAP's Enterprise AI Gateway?

It is a new SAP layer that consolidates authentication, policy enforcement, routing, monitoring, usage tracking and auditing for agents; SAP plans general availability in the first quarter of 2027.

Why do AI agents need their own identities?

Agents using shared or borrowed credentials cannot be traced to an owner or limited to the access they need. Individual identities make agents auditable and allow credentials to be revoked in real time.

Filed under: AI Governance & Trust, Agentic AI, Security & Fraud