IBM has announced new AI agent identity, security and governance capabilities at TechXchange 2026, aiming to help enterprises control how autonomous AI systems access business applications, infrastructure and sensitive data. The updates span HashiCorp Vault, IBM watsonx Orchestrate, IBM Integration Intelligence Hub, IBM Concert and Terraform, bringing together agent discovery, short-lived credentials, runtime isolation, performance evaluation and governed infrastructure changes. The goal is to make AI agents easier to identify, monitor and control as organisations deploy them across hybrid cloud environments and third-party platforms.
IBM Targets Enterprise AI Agent Risks With Identity and Control Tools
IBM is expanding its enterprise AI portfolio with a set of capabilities designed to address one of the biggest obstacles to agentic AI adoption: controlling autonomous software that can access systems, invoke tools and perform business actions.
Announced at TechXchange 2026 on October 27, the updates bring together identity management, credential security, agent discovery, runtime containment and performance evaluation. The portfolio spans IBM watsonx Orchestrate, HashiCorp Vault, IBM Integration Intelligence Hub, IBM Concert and Terraform, alongside an integration with NVIDIA OpenShell.
The strategy reflects a shift in enterprise AI infrastructure. As companies move beyond conversational assistants towards agents that execute multistep workflows, security teams need to know which agents exist, who owns them, what permissions they hold and how their actions can be audited.
AI agent governance becomes an infrastructure problem
The scale of the challenge could grow quickly. Gartner predicted in April 2026 that the average global Fortune 500 enterprise would have more than 150,000 AI agents in use by 2028, up from fewer than 15 in 2025. The research also found that only 13% of organisations believed they had appropriate AI agent governance in place.
IBM’s response is to treat agents as managed enterprise identities rather than temporary software processes. Its expanded HashiCorp Vault capabilities use short-lived credentials, centralised secrets management and least-privilege access controls to limit the resources an agent can reach.
Agent Identity, when used with Vault, provides a record of an agent’s ownership, lifecycle status and authorised scope. IBM says access can be denied and sessions revoked when an agent attempts an unauthorised action. This is intended to reduce the exposure created by persistent credentials and excessive permissions.
Vault 2.2 also introduces secured agent accounts, automatic credential renewal and post-quantum encryption. Secure Hub, meanwhile, is in public preview and provides centralised visibility across Vault environments.
These capabilities address a critical distinction in AI security: authenticating an agent does not automatically make every action it takes safe. Organisations must also constrain its authority and be able to withdraw access when conditions change.
Watsonx Orchestrate extends visibility across AI platforms
IBM is also expanding watsonx Orchestrate beyond agents built within its own ecosystem. The platform’s new Discover capability, currently in tech preview, connects with environments including Amazon Bedrock, Microsoft Foundry and Google Gemini to identify agents, models and Model Context Protocol (MCP) servers across third-party platforms.
Contain, also in tech preview, runs agents in isolated sandboxes and adds a kill switch for shutting down an agent. The Evaluate capability supports testing and monitoring agent quality, accuracy and compliance, alongside tracking model and token usage to help teams understand operating costs.
Together, these functions target what enterprises increasingly face as teams develop agents independently: fragmented inventories, inconsistent security policies and limited insight into production behaviour.
The approach also creates a competitive dimension. Microsoft, Google and Amazon provide their own AI development and cloud ecosystems, while IBM is positioning watsonx Orchestrate as a cross-platform management layer. Whether that strategy succeeds will depend on the depth of integrations and how consistently controls work across different agent frameworks.
Connecting agents to existing systems
IBM Integration Intelligence Hub adds a business-context layer through its Unified Context Engine. The product is designed to help agents understand existing APIs, system dependencies and reusable capabilities before carrying out tasks such as troubleshooting stuck messages or deploying APIs.
The objective is to let AI agents work with existing enterprise infrastructure rather than requiring organisations to rebuild applications around each new AI workflow. Skills, approval policies and audit controls are intended to constrain those actions.
IBM is also integrating Agent Identity and Vault with NVIDIA OpenShell, a secure runtime for AI agents. The combination is designed to pair an isolated execution environment with verified identities and short-lived credentials, helping organisations trace and limit agent activity.
The integration is significant because runtime isolation and identity controls solve different problems. A sandbox can limit interference between agents, while credential controls govern what an agent can access outside that environment. Neither eliminates the need to test agent behaviour or review consequential decisions.
Infrastructure security and legacy systems remain in scope
IBM’s announcement extends beyond agent management into infrastructure operations. IBM Concert combines observability, incident management, security and optimisation to help teams identify vulnerabilities and assess their operational impact.
Terraform adds governed infrastructure-change workflows through capabilities including Terraform policy, a Developer Portal and a remote MCP server. Availability varies: Terraform policy is generally available, while the Developer Portal and remote MCP server are in public preview.
IBM Z Atlas, in tech preview, maps dependencies across z/OS infrastructure, middleware and applications. The tool is designed to help teams assess proposed changes and validate their effects before implementation, reducing the risks associated with complex legacy-system modifications.
The wider portfolio reflects IBM’s view that enterprise AI depends on more than models and agents. Data context, secure identities, infrastructure visibility and controlled change management are also necessary to put autonomous systems into production.
What enterprises should evaluate
Some announced capabilities are generally available, while others remain in public or technical preview. Buyers should distinguish production-ready functions from features still being tested and confirm which integrations are supported in their own environments.
Enterprises evaluating the portfolio should test credential revocation, agent discovery coverage, audit trails, sandbox isolation and policy enforcement under realistic conditions. They should also measure false positives, agent task accuracy, operating costs and the extent of human approval required for sensitive actions.
IBM’s announcement offers a broad framework for governing agentic AI across hybrid infrastructure. Its practical value will depend on how reliably these controls operate across heterogeneous systems and whether security teams can manage agents without creating new operational complexity.
Market Landscape
Enterprise AI security is evolving from traditional identity and access management towards dedicated controls for autonomous agents, including runtime isolation, credential lifecycle management, evaluation and continuous monitoring.
- Agent proliferation: Gartner forecasts more than 150,000 agents at the average global Fortune 500 enterprise by 2028, while only 13% of organisations believe their governance is adequate.
- Identity-first security: IBM extends HashiCorp Vault’s secrets management and credential controls to AI agents, addressing risks from long-lived credentials and excessive permissions.
- Cross-platform management: watsonx Orchestrate aims to discover and evaluate agents across IBM, Amazon, Microsoft and Google environments.
- Infrastructure integration: IBM Integration Intelligence Hub, Concert and Terraform connect agent governance with enterprise applications, security operations and infrastructure changes.
- Competitive differentiation: IBM is pursuing a hybrid, cross-platform strategy rather than limiting agent management to a single AI development ecosystem.
Top Insights
- IBM combines agent identity, short-lived credentials, access control and runtime isolation to address security risks as autonomous AI systems enter enterprise workflows.
- watsonx Orchestrate adds cross-platform agent discovery, containment and evaluation, including visibility into model usage and token-related operating costs.
- HashiCorp Vault and Agent Identity aim to enforce least-privilege access and support rapid credential revocation when agents exceed authorised boundaries.
- NVIDIA OpenShell integration combines protected agent execution with verified identities and scoped credentials for more controlled enterprise AI operations.
- Availability varies across the portfolio, making production readiness, supported integrations and measurable security outcomes essential evaluation criteria for enterprise buyers.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
