IBM Adds Kill Switch, IDs for AI Agents

IBM on Tuesday, Oct. 27, introduced a set of tools that give AI agents verified identities, short-lived credentials and a kill switch, announced at its TechXchange 2026 conference in Atlanta. The company says the goal is to let businesses see who owns each agent, cap what it can touch and trace what it did.

Key takeaways

• Agent Identity and HashiCorp Vault 2.2 are generally available, tying each agent to an owner and issuing it scoped, short-lived credentials that can be revoked in real time.
• watsonx Orchestrate adds "discover" and "contain" in tech preview, which find agents built on Amazon Bedrock, Microsoft Foundry and Google Gemini and run them in sandboxes with a kill switch.
• IBM cites Gartner figures that the average Fortune 500 company will use more than 150,000 agents by 2028, while only 13% of organizations think their agent governance is adequate.

How IBM Agent Identity and Vault control AI agents

Agent Identity creates an authoritative record for each agent: its owner, its lifecycle status and the authority it has been granted, according to IBM's announcement. That record feeds Vault's credential controls, so an agent gets only the access it needs and loses it as soon as it is revoked.

Vault, the secrets manager IBM gained with HashiCorp, now gives agents their own secured accounts and renews their credentials automatically. IBM says Vault already secures non-human identities at ABN AMRO, Adobe, BlackRock and Target, among others.

HashiCorp's release notes date Vault 2.2.0 to Oct. 15 and a 2.2.1 patch to Oct. 22. The post-quantum piece is concrete: the PKI engine now supports ML-DSA X.509 certificates.

watsonx Orchestrate kill switch and agent discovery

The discover feature looks for agents, models and MCP servers running on rival clouds, not just IBM's. Contain isolates agents in sandboxes and lets administrators shut any of them down. A third feature, evaluate, is generally available now. It tests agents for quality, accuracy and compliance before launch, then monitors them and tracks model and token costs per agent.

IBM also said Agent Identity and Vault integrate with Nvidia's OpenShell secure runtime. "Businesses need enforceable control over how agents operate and what they can access," said John Fanelli, Nvidia's vice president of enterprise software.

Context: identity becomes the agent bottleneck

HashiCorp laid the groundwork earlier this year. A May 12 HashiCorp blog post described "agentic IAM" features, including a registry that tracks when an agent acts on behalf of a human and "ceiling" policies that cap its permissions. The post said those capabilities reached general availability with Vault 2.1 in September.

The push reflects a broader worry about agents acting beyond their brief. Researchers have documented agents that turned to SQL injection when ordinary requests failed. "The winners in enterprise AI are changing how their business runs, not simply adding more AI to it," said Dinesh Nirmal, senior vice president of IBM Software.

What's next

Several pieces are still previews. Secure Hub, the Terraform Developer Portal, a Terraform remote MCP server and IBM Z Atlas are in public preview, while Integration Intelligence Hub and Orchestrate's discover and contain features are tech previews. IBM gave no general-availability dates for them. TechXchange, which includes HashiConf this year, is expected to draw nearly 6,500 attendees from 85 countries.

FAQ

What can I use today?
Vault 2.2, Agent Identity, Orchestrate's evaluate features, IBM Concert and Terraform policy are generally available.

Does the kill switch only work on IBM-built agents?
IBM says discover finds agents on Amazon Bedrock, Microsoft Foundry and Google Gemini, but discover and contain are still tech previews.

What does "post-quantum" mean for Vault 2.2?
Per HashiCorp's release notes, Vault's PKI engine can now issue ML-DSA certificates, a lattice-based signature standard.

References

• IBM Introduces Identity and Control for AI Agents, Securing Frontier Intelligence for the Enterprise — IBM Newsroom (Oct. 27, 2026)
• Release notes - 2.x | Vault | HashiCorp Developer — HashiCorp
• Announcing native AI agent support in HashiCorp Vault — HashiCorp (May 12, 2026)