IBM gave AI agents identity and control: the agent is now governed like an employee
Direct answer
On October 27, 2026, at TechXchange 2026 in Atlanta, IBM announced new capabilities so developers, IT and security teams can put AI agents to work across business systems with clear authority and accountability. The proposal brings four pieces together: agent discovery, verified identity, access controls and performance evaluation, to answer who owns each agent, what it can touch and what it did afterwards.
The pieces, with their availability status
- Agent Identity (generally available): an authoritative record per agent with its designated owner, its lifecycle phase and the authority delegated for work across systems. When an agent attempts an action outside its permissions, the system can deny access and revoke the session in real time.
- HashiCorp Vault 2.2 (generally available): gives each agent its own secured account, renews its credentials automatically and applies least-privilege access with short-lived credentials. The announcement also mentions post-quantum encryption for quantum-era threats.
- Evaluate in watsonx Orchestrate (generally available): tests the quality, accuracy and compliance of agents, models and MCP servers before deployment, then monitors them in production. Built-in cost tracking shows model and token use per agent.
- Discover and Contain (tech preview): Discover connects to Amazon Bedrock, Microsoft Foundry and Google Gemini to find agents, models and MCP servers built on other platforms. Contain runs agents in isolated sandboxes, so a problem agent cannot interfere with others, and offers a kill switch.
Why this matters beyond IBM
- Inventory comes first: a company cannot govern agents it does not know exist. Cross-platform discovery answers the most common 2026 case: agents created by several teams on several clouds, without a central record.
- Identity is the permission: treating the agent like an employee with its own revocable credential changes the security conversation. The question stops being which model it uses and becomes which credential it holds, for how long, and who can revoke it.
- The figure IBM cites: IBM cites Gartner in noting that the average Fortune 500 enterprise could use more than 150,000 agents by 2028. That is a Gartner projection cited in the announcement, not an IBM measurement, and it sizes the governance problem rather than promising that volume for any specific business.
- The announcement example: IBM names the Albert Einstein College of Medicine applying Verify and Vault to agent identity, just-in-time access, dynamic secrets and auditability. It is an announcement example, not a guaranteed outcome for another sector.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- Island launched an enterprise browser for AI agents
- Mintive: every agent with its own credential
- Privacy and data: what happens to your customers' information
Frequently asked questions
What did IBM announce on October 27, 2026?
New TechXchange capabilities to govern AI agents: verified identity, Vault 2.2 with short-lived credentials, and watsonx Orchestrate with discovery, containment and evaluation.
What is available now and what is still in preview?
Agent Identity, HashiCorp Vault 2.2 and Evaluate are generally available. Discover and Contain are in tech preview.
What does an agent identity record?
Its designated owner, its lifecycle phase and its delegated authority, with real-time session revocation for an unauthorized action.
Does this apply only to agents built with IBM?
Not per the announcement: Orchestrate works across platforms and Discover connects to Amazon Bedrock, Microsoft Foundry and Google Gemini.
