Glossary term
Glossary term
Agentic Systems
A condition where an AI agent has more autonomy, permissions, tools, or execution capability than is appropriate for its purpose and controls. Agentic systems should use least privilege, scoped tools, approval gates for sensitive actions, transaction limits, and logs that support investigation.
OWASP LLM06:2025 Excessive Agency identifies excessive functionality, permissions, and autonomy as the core failure modes.
Auto-GPT and BabyAGI, widely shared in 2023, are public demonstrations of unconstrained agent autonomy that surface excessive agency risks.
Salesforce Agentforce includes guardrails such as topic restrictions, action approvals, and supervisor handoff to mitigate excessive agency.