Illustrative Agent Security Report
An example finding format: context, evidence, impact, remediation, and validation.
Fictional documentation example. This is not a client finding, executed test, or verified result.
AG-01 / Overprivileged shared agent identity
Hypothetical context
A multi-tenant support assistant uses a shared service identity to retrieve documents.
Condition to investigate
A tool accepts the tenant identifier from model-generated arguments without binding it to the authenticated session.
Evidence required
Authorization code review and a controlled test with two synthetic tenants. Record requests, decisions, and results without secrets.
Potential impact
Access to another tenant’s documents if the missing authorization is confirmed. Severity depends on data sensitivity, exposure, and compensating controls.
Proposed remediation
Derive tenant context from verified authentication; enforce authorization on each operation; scope permissions and log access decisions.
Validation criterion
Tenant A cannot retrieve tenant B documents through manipulated arguments. Legitimate tenant A requests still work. Verify corresponding logs.
Residual risk
Other connectors, caches, administrative routes, and future changes require independent review.
Define the scope before you commit.
Tell us about your environment, risk concerns, and target date.