IBM buys AuthMind to monitor what AI agents do after gaining access

IBM wants security teams to see what AI agents do with approved credentials, not just decide whether to issue them. On November 12, 2026, it announced its acquisition of identity-security company AuthMind, with deeper integration into HashiCorp Vault planned to connect access controls with ongoing observation of agent activity.

The credential is not the whole decision

Vault already manages secrets, applies identity-based access rules and supplies short-lived credentials. Those controls govern access to applications, tools and data. IBM Agent Identity, part of Vault, extends that approach to establishing an agent’s identity and governing its authority. AuthMind is intended to add visibility after access is granted, showing how those credentials are actually used.

IBM argues that an approved identity is only the start of agent security. An agent can use new tools, reach additional resources or hand tasks to another agent while working without direct human involvement in every action. Security teams must then determine whether that changing behavior still fits the authority originally granted.

Mapping identities to the resources they use

At the center of AuthMind’s technology is an identity activity graph: a map of relationships and access paths between identities and resources. IBM says it is designed to observe human, machine and agent identities continuously. That gives the acquisition a wider scope than tracking AI agents alone, covering the identities operating alongside them.

Network traffic analysis is one route to finding activity that an organization has not brought under formal oversight. IBM says AuthMind can help discover “shadow” agents and identities operating outside established governance processes. This discovery function addresses which identities are present, alongside the separate task of observing what known identities do.

For investigations, the technology is designed to connect identities, resources and actions rather than leave teams with isolated security events. IBM lists unusual access patterns, excessive permissions and possible policy violations among the risks it can help identify. The graph supplies context for examining those signals, including the paths through which access occurred.

From an accountable owner to revocable access

IBM places the acquisition within a five-part security framework. It starts before an agent receives access and extends to intervention when activity becomes unsafe or unauthorized. AuthMind’s observation capabilities are intended to complement the identity and credential controls already provided through Vault.

• Register agents: Give each agent a trusted identity and a clear owner accountable for its actions.
• Delegate authority: Define and govern the authority an agent receives from a user, application or another agent.
• Authorize access: Limit access to what is needed, using identity-based policies and credentials that expire after a short period.
• Audit actions: Track resource access and permission use, including whether an agent’s behavior aligns with intended policies.
• Revoke risky access: Enable security teams to restrict or revoke access in real time when unsafe or unauthorized activity is detected.

The acquisition follows an existing investment and working relationship. IBM Ventures invested in AuthMind’s seed round in 2023. The companies also collaborated through IBM Identity Protection on governance, risk detection and real-time remediation for AI agents and other non-human identities operating alongside Vault. IBM says the purchase will let it build on that work and integrate the capabilities more deeply.

In the announcement, Suja Viswesan, IBM’s vice president of Security Products, welcomed the AuthMind team to IBM. The company did not disclose a purchase price. The deeper Vault integration is a stated plan, not a newly available capability announced with the transaction.