The collaboration between Yubico and Delinea highlights a critical PAM question for the agentic era: how can an organisation prove that an automated action was authorised by the right human, performed by the right agent, and limited to the right privilege? Strong authentication and privileged access management address different parts of that accountability chain, but they become substantially more valuable when used together.
AI agents create an attribution gap. A human may approve a task, but the agent can interpret instructions, select tools, and perform multiple downstream actions without another explicit checkpoint. If the agent uses a shared credential or a standing administrative role, investigators may be unable to distinguish authorised automation from misuse. That ambiguity weakens privileged account security and makes post-incident analysis much harder.
Hardware-backed authentication can establish a stronger starting point for the human approval event. PAM then needs to bind that approval to a specific agent, target, action, and time window. The result is not blanket permission; it is a verifiable authorisation for a bounded operation. If the request changes materially, the policy should require a new approval rather than allowing the agent to continue under the original grant.
Session management is central to making this model operational. A privileged session record should show who approved the action, which agent executed it, what credentials or tokens were brokered, and which commands or API calls followed. Recording that chain lets security teams investigate the full transaction instead of relying on a single authentication event that says little about what the automation actually did.
There is also a resilience benefit. If a key is lost, an agent behaves unexpectedly, or an approval policy is changed, centrally managed privilege can be revoked without redeploying every workflow. Short-lived credentials, step-up authentication, and explicit human approval for high-impact actions reduce the blast radius while preserving automation for lower-risk tasks.
For CISOs, agentic accountability should be treated as a control objective across PAM and identity architecture. The best design will not force a human to approve every low-risk action, but it will make high-risk privilege explainable, attributable, and reviewable from request through execution.
Source: Business Wire