AI & data platforms · AI Platform · Identity Security · Application Owners
Govern credentials used by autonomous AI agents
Which agent instance may call which tool, with what credential, data boundary, and expiration?
Operational pain
Long-lived API keys and shared service accounts let an agent's prompt or plugin compromise become broad infrastructure access.
Trigger
Agent rollout, MCP integration, tool expansion, or credential misuse.
QNSI contribution
Connect the decision to a controlled security path
Inventory agent and tool identities, keys, scopes, algorithms, owners, environments, and rotation events in QNSI.
Decision artifact
An agent credential register with least-privilege scope, per-tool trust, session lifetime, revocation, and orphan detection.
What still requires validation
The operator tests authorization, prompt-injection containment, secret isolation, human approval, audit completeness, and shutdown.
External problem context
Primary sources
These sources establish the external requirement or risk context. They do not endorse HEOSSI or prove that QNSI completed this scenario.
Evidence boundary
What this page does—and does not—prove
This is a product evaluation pattern, not a customer case study, certification, legal opinion, regulator endorsement, or claim that a production deployment completed the described work.