AI News
IBM Gives AI Agents Verified Identities, Least-Privilege Access and Real-Time Revocation
At TechXchange 2026 in Atlanta, IBM announced on October 27 a set of capabilities meant to answer a simple question companies keep failing to answer: which agent did that, who owns it, and what was it allowed to touch? The announcement covers verified identities for agents, least-privilege access, and real-time revocation when an agent tries something unauthorised.
The scale of the problem is the pitch. IBM cites Gartner research forecasting that the average Fortune 500 enterprise will run more than 150,000 AI agents by 2028, while only 13% of organisations believe they have the right agent governance in place. The gaps IBM names are familiar to anyone running agents today: credentials that never expire, no clear owner per agent, and controls that do not follow an agent across teams, platforms and clouds.
What IBM actually announced
The announcement has two main parts. First, HashiCorp Vault, IBM's secrets-management product, is extended to AI agents. Vault issues short-lived credentials to agents instead of long-lived keys, gives each agent a dedicated account, renews credentials automatically, and adds post-quantum encryption. Vault 2.2 is the version carrying the agent features, and a new Secure Hub, in public preview, gives a central view across an enterprise's Vault deployments.
Second, IBM watsonx Orchestrate is positioned as the place that discovers and governs agents no matter where they were built, including agents running on Amazon Bedrock, Microsoft Foundry and Google Gemini. Its technical preview includes automated discovery across external models and servers, sandbox containment with a kill switch to halt a rogue process, and tooling to track model costs and token usage.
IBM frames all of this as an "AI operating model" of four systems working together: agents, data, automation and hybrid infrastructure. Dinesh Nirmal, IBM's Senior Vice President for Software, put it this way: for agents to take on real work across the systems a company depends on, they need "the same discipline as its most critical systems" — and trust and control built into the model, not bolted on afterwards.
Why this matters now
This is the enterprise version of a story already playing out elsewhere. SK Telecom recently gave its AI agents employee IDs, named supervisors and performance reviews, treating agents as a new form of labour that must be accounted for. IBM's announcement is the tooling version of the same idea: an agent that can spend money, move data and change systems needs an identity that can be checked, limited and switched off.
The revocation piece is the most important detail. An agent with a permanent API key is a standing risk: if it is hijacked, or if its instructions are poisoned, the key keeps working until someone notices. Short-lived credentials plus a kill switch change that equation, because a compromised agent loses its access quickly instead of keeping it indefinitely.
What to keep in perspective
Not everything announced is shipping today. Secure Hub is in public preview, and the watsonx Orchestrate discovery and sandbox features are in technical preview, which means they are early and may change. IBM did not publish pricing for the new capabilities in the announcement, and governance tools only help if companies actually use them: an identity system nobody enforces is no better than no system at all.
Still, the direction is clear. The agent market is moving from "let the agent do it" to "prove which agent did it, with what permission, and stop it when it goes wrong." IBM's bet is that the companies that win enterprise AI will be the ones that build that discipline first.