Least Agency

Least privilege answers which resources an identity may touch. Least agency adds a second question for autonomous systems: how much independent decision-making the agent should have. OWASP promotes the principle in its agentic security guidance.

Applying least agency involves:

• Minimal toolsets: Connecting only the tools a workflow actually needs.
• Task-scoped credentials: Issuing short-lived permissions per task rather than standing access.
• Bounded autonomy: Defining which actions an agent may take alone and which need approval.
• Constrained destinations: Allowlisting the systems, domains, and data an agent may reach.
• Regular review: Removing tools and permissions that agents stop using.

Least agency is a design discipline as much as a control. Teams often grant broad capabilities during prototyping and never revisit them, leaving production agents far more powerful than their jobs require.

Because no defense against prompt injection or model error is perfect, least agency ensures that when something goes wrong, the agent simply cannot do much damage.

How PointGuard AI Helps

PointGuard AI Agent Mission Control defines each agent's mission, tools, and allowed destinations, and enforces them on every action. AI Security Posture Management continuously identifies agents whose granted capabilities exceed observed needs, helping teams apply least agency across the fleet.

Learn More

• OWASP Top 10 for Agentic Applications
• OWASP LLM06:2025 Excessive Agency
• NIST AI Risk Management Framework