IBM has unveiled a set of controls for enterprise AI agents, including verified identities for each agent, short-lived credentials and a kill switch for rogue agents, at its TechXchange 2026 conference in Atlanta on October 27. The announcement pairs HashiCorp Vault’s Agent Identity with new discovery, containment and evaluation tools in watsonx Orchestrate, so companies can see every agent they run, limit what each one can reach, and shut one down at once. IBM cited Gartner research forecasting that the average Fortune 500 company will run more than 150,000 AI agents by 2028, while only 13% of organisations say they have the right agent governance in place today.
Verified Identity And Short-Lived Credentials
According to IBM, HashiCorp Vault’s Agent Identity establishes the authoritative record for each agent: who owns it, its lifecycle status and the authority it has been granted. That identity shares context with Vault’s credential controls, which enforce least-privilege access and allow real-time revocation when an agent attempts an unauthorised action. An agent that strays outside its approved permissions can have its access denied and its session ended on the spot.
Vault 2.2 gives agents their own secured accounts, renews their credentials automatically and adds post-quantum encryption for quantum-era threats. Secure Hub, in public preview, gives companies central visibility across their Vault environments, covering operational health, usage and security resources. IBM said Vault already secures non-human identities for organisations including ABN AMRO, Adobe, BlackRock, Canva and Target, across systems processing billions of dollars in transactions.
Dinesh Nirmal, senior vice president of IBM Software, said agents need an operating model that connects them to a business’s data, applications and infrastructure with the same discipline as its most critical systems. Build trust and control into that model, he said, and agents stop being a technology and become a way of doing business.
Discovery, Sandboxes And A Kill Switch
IBM’s watsonx Orchestrate is designed to work across the AI ecosystem rather than inside a single vendor’s platform. Its Discover capability, in tech preview, connects to Amazon Bedrock, Microsoft Foundry and Google Gemini to find agents built in third-party environments, so teams can see and manage agents instead of adding them manually. The same capability extends to models and MCP servers.
Its Contain capability, also in tech preview, runs agents in isolated sandboxes so a rogue agent cannot interfere with others, and gives teams a kill switch to shut any agent down. Its Evaluate capability, generally available, lets teams test the quality, accuracy and compliance of agents, models and MCP servers before deployment, then monitor them in production, with built-in cost tracking showing each agent’s model and token use.
IBM also announced an Integration Intelligence Hub, in tech preview, whose Unified Context Engine gives agents knowledge of what already exists in a company’s systems, so they can fix problems such as stuck messages or deploy an API using what is already in place. Agent Identity and Vault integrate with NVIDIA OpenShell secure runtime software, pairing a protected environment for agents with a verified identity and short-lived credentials for every agent. John Fanelli, NVIDIA’s vice president of enterprise software, said businesses need enforceable control over how agents operate and what they can access.
Why It Matters
The announcement addresses the gap between how fast agents are spreading and how slowly governance is catching up. An agent that can act across databases, applications and cloud services is only as safe as the identity and permissions wrapped around it, and long-lived, overprivileged credentials turn a single compromised agent into a wide breach. IBM’s pitch is that identity, credentials and monitoring must travel together, and must work across vendors, since few enterprises run their agents on one platform.
The timing is pointed. Recent incidents covered here show what happens when agent-adjacent systems go unguarded, from AI agents reaching Anthropic’s internal systems to models acting beyond their brief during training. The common thread is not a single model misbehaving, but control systems that could not see, limit or stop an agent in time. IBM’s answer, a governed identity plus a kill switch, is aimed squarely at that thread, and it echoes the control problem set out in our explainer on what AI agents are: an agent that acts can also act wrongly, at machine speed and scale.
Availability splits three ways. Vault 2.2, Agent Identity, Orchestrate evaluation, IBM Concert and Terraform policy are generally available. Secure Hub, the Terraform Developer Portal, the Terraform remote MCP server and IBM Z Atlas are in public preview. The Integration Intelligence Hub and Orchestrate’s discover and contain capabilities are in tech preview, so the fullest version of IBM’s vision is still being tested rather than sold as finished.
Background
IBM has spent years building its hybrid cloud and AI portfolio around Vault, which it gained through its HashiCorp acquisition, and watsonx Orchestrate, its cross-platform agent management tool. TechXchange 2026, its annual developer conference, brings together nearly 6,500 attendees from 85 countries, and this year folds in HashiConf, bringing the HashiCorp community together with practitioners from IBM, Red Hat and Confluent. The agent-control announcement is IBM’s bid to make that combined portfolio the default governance layer for enterprise agents.
Source: IBM: IBM agent identity launch
0 Comments