AI agent identity security arrives as IBM gives every agent an owner and expiring credentials

AI agent identity security stopped being a conference talking point on 27 October 2026. At TechXchange 2026 in Atlanta, IBM announced new capabilities to give software agents verified identities, limited credentials and a named owner, according to IBM’s announcement via PR Newswire. The pitch is plain: you can’t scale agents you can’t see, test or account for.

Information verified through 27 October 2026.

Agents are no longer demos. They read databases, call tools, message other agents and spend money on tokens. Each one acts, in effect, like a member of staff nobody hired through HR. IBM’s answer is to treat them that way — badge, manager, restricted keys, dismissal procedure included.

Why AI agent identity security became an enterprise problem: the numbers behind how to govern AI agents in the enterprise

IBM cited Gartner research that by 2028 the average Fortune 500 enterprise will use over 150,000 AI agents. Only 13% of organisations believe they have the right AI agent governance in place, on the same research, as Unite.AI summarised. Gartner’s underlying release, dated 28 April 2026, put the 2025 baseline at fewer than 15 agents per enterprise. From fewer than 15 to 150,000 in three years is not growth. It’s a flood.

The blockers IBM names are familiar to any security team: persistent credentials, unclear ownership and disconnected controls. An agent holding a long-lived, wide-open key is a breach waiting for a trigger. If it is compromised, or simply wanders outside its task, that key opens every door it was ever handed.

That is the practical core of how to govern AI agents in the enterprise. Not a policy PDF. A record of who owns each agent, what it may touch, and a way to cut it off mid-action.

HashiCorp Vault AI agent credentials that expire

The credential half of the launch runs through HashiCorp Vault AI agent credentials. Vault already manages secrets for applications and services — centrally held credentials, automated rotation and dynamically generated credentials that expire or can be revoked, IBM said. Its existing non-human identity customers include ABN AMRO, Adobe, BlackRock, Canva, Starbucks and Target, on IBM’s list.

Agent Identity, now generally available, establishes the authoritative record for each agent: who owns it, its lifecycle status and the authority it has been granted, the announcement said. Paired with Vault, it enforces least-privilege access. If an agent tries to retrieve data or act outside its approved permissions, access can be denied and the session revoked in real time.

Vault 2.2 gives agents their own secured accounts, renews their credentials automatically and adds post-quantum encryption. Secure Hub, in public preview, gives companies a single view across their Vault deployments. Short-lived credentials shrink the window in which a stolen key is worth anything. That won’t stop a bad agent. It limits how far a bad agent gets before its keys die in its hands.

At TechXchange, Albert Einstein College of Medicine’s Shailesh M. Shenoy, assistant dean for information technology, said technologies such as IBM Verify and Vault can help organisations establish the identity, access and security controls needed to innovate with confidence, in IBM’s release. Client speakers from athenahealth, Optum, Seven-Eleven Japan and Yum! Brands were also listed on Vault-related work.

IBM watsonx Orchestrate agent discovery across rival platforms

Identity solves half the problem. You still have to find the agents. IBM watsonx Orchestrate agent discovery, in tech preview, connects to Amazon Bedrock, Microsoft Foundry and Google Gemini to find agents built in third-party environments, IBM said — including models and MCP servers, not just IBM-built agents. RuntimeWire noted that this positions Orchestrate as a control plane across competing cloud platforms, well beyond IBM’s own tools.

Containment, also in tech preview, runs agents in isolated sandboxes so one rogue agent can’t interfere with others, with a kill switch to shut any agent down. The evaluation tools, generally available, let teams test quality, accuracy and compliance before deployment, then monitor agents in production, with cost tracking for model and token use, El Fondo reported.

IBM also tied the stack to NVIDIA OpenShell, pairing a policy-governed runtime with verified identity and short-lived credentials. NVIDIA’s John Fanelli said businesses need enforceable control over how agents operate and what they can access, in IBM’s release. Dinesh Nirmal, IBM’s senior vice president for software, framed the whole launch as an operating model: agents taking on real work need the same discipline as a company’s most critical systems.

Availability is staggered, and buyers should read it carefully. Vault 2.2, Agent Identity and Orchestrate’s evaluate capabilities are generally available. Secure Hub is in public preview. Discovery and containment are in tech preview — announced, demonstrated, not yet general release.

Analysis: IBM is not selling a smarter agent here. It’s selling the staff file, the key cabinet and the off switch for everyone else’s agents, including agents built on Amazon, Microsoft and Google platforms. That’s a shrewd place to stand if Gartner’s 150,000-agent forecast is even half right, because governance spend follows deployment whether the agents work well or not. The risk is equally clear. Discovery and containment — the two capabilities that answer “what is running, and can we stop it?” — are the two still in tech preview. Until they ship generally, enterprises get strong credentials for the agents they know about, and a promise about the ones they don’t.

Sources

1. https://www.morningstar.com/news/pr-newswire/20261027ny81742/ibm-introduces-identity-and-control-for-ai-agents-securing-frontier-intelligence-for-the-enterprise
2. https://runtimewire.com/article/ibm-agent-identity-vault-watsonx-orchestrate
3. https://www.el-fondo.com/en/news/ibm-launches-security-tools-for-ai-agents-across-clouds
4. https://www.unite.ai/ibm-brings-governed-identity-and-credential-controls-to-ai-agents/