Search the Atlas

Search risks, controls, and glossary terms

HighAgenticPrivilege CreepDemonstrated

Silent Permission Expansion

Agent Autonomy

Description

Agent permissions expand incrementally over time through implicit grants, context inheritance, or unreviewed requests. Accumulated privileges far exceed original mandate—a slow-burn access-control failure.

Example scenario

Over three months an HR agent accumulates access to payroll, benefits, and personnel records through incremental grants.

Real-world evidenceDemonstrated

Research on multi-agent orchestration and tool-augmented LLMs has shown that agents can inherit and exploit overly broad OAuth scopes, context-passed credentials, and implicit permissions granted by upstream agents. This has been demonstrated in controlled experiments and published red-team reports, though systematic silent expansion in production enterprise IAM is not yet a confirmed incident.

Primary mitigations

  • Regular privilege audits
  • time-bound permission grants
  • explicit re-authorisation for new capabilities
  • minimum-necessary access policy.

Detection signals

Privilege Expansion Rate; permission delta per session.

Mitigating controls

6
Dual coverage

Related risks in Agent Autonomy