On October 27, 2026, at TechXchange 2026 in Atlanta, IBM announced identity and control capabilities for AI agents running inside enterprises: every agent should have a verifiable identity, receive only the minimum permissions its task requires, and lose its credentials in real time if it oversteps. The backdrop is a Gartner forecast that by 2028 the average Fortune 500 company will run more than 150,000 agents, while only 13% of organizations believe they have the right agent governance in place.
Vault manages identity; credentials are short-lived only
The core is the combination of HashiCorp Vault and Agent Identity. Agent Identity records each agent's owner, lifecycle status and granted authority, forming an authoritative record; Vault then issues short-lived credentials based on that record, so they stop working when they expire or are revoked, instead of leaving a long-lived super-credential that can be abused after an agent is compromised. The newly released Vault 2.2 also gives agents their own secured accounts, renews credentials automatically and adds post-quantum encryption. IBM also launched Secure Hub in public preview for centralized visibility across Vault environments.
Visible, containable and measurable
IBM watsonx Orchestrate handles discovery and control across platforms: its discovery capability connects to Amazon Bedrock, Microsoft Foundry and Google Gemini to find agents, models and MCP servers scattered across environments; containment runs agents in isolated sandboxes with a kill switch; evaluation tests quality, accuracy and compliance before deployment, then keeps monitoring in production and tracks each agent's model and token spend. Note the status difference: discovery and containment are in tech preview, while the evaluation capabilities, Vault 2.2 and Agent Identity are generally available, so preview features should not be bought as finished products.
For companies already running agents on several platforms, the value of this combination is bringing "which agents are running, what they are doing and what they can touch" under one set of controls instead of leaving each platform to govern itself. It does not solve a more basic problem, though: identity and credential controls only bind agents that have been brought under management, while externally connected agents adopted informally by employees remain out of sight, so the reach of governance depends on the company first knowing what it actually has.
