IBM is assembling a set of identity, credential, and orchestration tools designed to help enterprises discover, identify, manage, and control the fast-growing population of AI agents operating inside their environments.

The package combines IBM Agent Identity, now available, with HashiCorp Vault and watsonx Orchestrate. IBM says the combination can identify and manage AI agents built across multiple platforms, giving security and IT teams a central way to understand which agents exist, who owns them, what they are allowed to do, and when their access should be revoked.

Key facts

• IBM is packaging IBM Agent Identity, HashiCorp Vault, and watsonx Orchestrate into an enterprise control approach for AI agents.
• The tools are intended to discover, identify, manage, and control agents built across multiple platforms.
• IBM Agent Identity registers agents in a central inventory, captures ownership and metadata, classifies risk, documents approved tool connections, and serves as an authoritative record for each agent.
• HashiCorp Vault is being extended to issue short-lived credentials to AI agents so each agent can perform only the actions it was assigned.
• IBM says the controls can enforce least-privilege access and support real-time revocation when an agent attempts an unauthorized action.
• watsonx Orchestrate works with Vault credential controls and Agent Identity to determine each agent’s job, govern its authority, and enforce least-privilege credential access.
• IBM has integrated HashiCorp Vault with Nvidia OpenShell secure runtime software to help enterprises scale agents.
• IBM describes the combined capabilities as a coordinated agentic identity security control plane spanning orchestration runtime, identity and access management, and secrets and credentials.

Why agent sprawl is becoming an enterprise problem

AI agents are moving quickly from experimentation into production workflows. They can interact with business systems, call APIs, retrieve data, trigger transactions, and act on behalf of users or business units. That autonomy is valuable, but it also creates a new class of non-human identity that must be governed with the same rigor as employees, applications, and service accounts.

The challenge is scale. Research firm Gartner predicts that by 2028, an average global Fortune 500 enterprise will have more than 150,000 agents in use, up from less than 15 in 2025. That trajectory points to significant agent sprawl, IT complexity, and management challenges. Agents can be created by business users, developers, data teams, and third-party platforms, often without centralized visibility.

IBM is positioning its package as a response to that fragmentation. The company says enterprises need a way to see every agent, understand what it is doing, and control its access without relying on security expertise from the business users who create or use the agents.

The visibility, audit, and revocation gap

Tyler Lynch, outbound CTO for IBM Automation, said the agentic challenges IBM wants to address are varied. In a press briefing, he described a basic visibility problem: anyone can launch an AI agent, but enterprises need to know which agents exist and what each one is doing.

Lynch also pointed to a security expertise gap. Business users and knowledge workers who create or use agents are not necessarily security experts. They may not understand least privilege, and they often operate under the principle of the user and simply impersonate that user. That creates an audit problem because it can be unclear whether an agent or a person performed a specific action or operation.

Those issues compound into governance and revocation problems. If an agent goes rogue or spends too much money, enterprises need to identify the behavior and stop it in real time. Traditional identity controls were not designed for a landscape in which large numbers of autonomous software agents can be created across platforms and act with delegated authority.

Shadow AI and the limits of blocking agents

Some organizations respond to agent risk by blocking or restricting agent use. Gartner has warned that this is not a long-term solution. If employees cannot work in sanctioned tools, they are likely to go around organizational controls and use shadow AI, which can present far greater risks.

Max Goss, senior director analyst at Gartner, has stated that organizations need to find a balance where they can govern agents and manage sprawl while also safely empowering employees to innovate with these tools. That balance requires visibility and control mechanisms that do not depend solely on prohibitions.

IBM’s approach is to make agents identifiable and governable rather than invisible. The goal is to let enterprises apply identity, credential, and policy controls to agents in a way that supports adoption while reducing unmanaged risk.

Extending Vault to AI agents

HashiCorp Vault currently manages the secrets and credentials that applications and services use to access business systems. It provides control through centrally managed secrets, automated rotation for supported credentials, and dynamically generated credentials that expire or can be revoked.

Vault already secures non-human identities for organizations including ABN AMRO, Adobe, athenahealth, BBVA, BlackRock, Canva, Duke Energy, Expedia, London Stock Exchange, Starbucks, and Swiss Federal Railways. IBM is now extending those capabilities to AI agents by issuing secured, short-lived credentials that limit each agent to what it needs.

That short-lived credential model is important because it reduces the blast radius of a compromised or misbehaving agent. Instead of giving an agent long-lived access that can be reused or leaked, Vault can issue credentials that expire quickly and can be revoked when risk changes.

Agent Identity as a central inventory

When used with HashiCorp Vault, IBM Agent Identity registers agents in a central inventory. It captures ownership and metadata, classifies risk, documents approved tool connections, and becomes the authoritative record for each agent, including who owns it, its lifecycle status, and the authority it has been granted.

Agent Identity is built to extend existing identity and security investments. IBM says it supports identity providers such as Okta, Microsoft Entra, and Ping to act as an agent registry. It can then share relevant identity and authorization context with Vault’s credential controls, enforcing least-privilege access and enabling real-time revocation when an agent attempts an unauthorized action.

Once an agent is known, IBM Agent Identity helps control how it acts. That means enterprises can move from ad hoc agent creation to a model in which each agent has an owner, a defined purpose, an approved set of connections, and a lifecycle that can be managed.

Orchestration, runtime security, and a coordinated control plane

IBM watsonx Orchestrate works with Vault’s credential controls and Agent Identity to ascertain what each agent’s job is, govern its authority, and enforce least-privilege credential access. Together, these controls help organizations authorize and audit agent actions and rapidly revoke trust and credentials when risk changes.

IBM has also integrated HashiCorp Vault with Nvidia OpenShell secure runtime software to help enterprises scale agents. The combination is intended to bring credential controls into the runtime environment where agents operate.

IBM describes the combined capabilities as a coordinated agentic identity security control plane spanning three domains where agent security is enforced: the agentic orchestration runtime, the identity and access management layer, and the secrets and credentials layer. The architecture is meant to connect agent discovery, ownership, authorization, credential issuance, and revocation rather than treating each as a separate silo.

For enterprises, the operational implication is that agent governance must become part of the broader identity and security fabric. Agents need owners, inventories, risk classifications, approved connections, short-lived credentials, audit trails, and revocation paths. As agent populations grow, the organizations that can apply those controls consistently across platforms will be better positioned to capture the benefits of agentic automation without losing visibility or control. The combined controls are intended to give enterprises a way to keep agent adoption from outrunning the identity and security systems that are supposed to govern it.

Source:Network World News