Excessive Agency
A condition in which an AI system is granted or exercises capabilities, permissions, autonomous decision-making authority, or ability to affect external systems beyond what is necessary for a task and its risks.
Example
A system asked only to present options also has permission to transmit information externally without first obtaining the user's confirmation.
Distinctions and cautions
AI Agent describes a system that may interact with tools and its environment across multiple steps. Excessive Agency asks whether the scope of permissions or external effects exceeds what is needed. It differs from Goal Misalignment and Unsafe Compliance, and does not describe the model as a person with human-like intentions.
Terminology note
Whether an AI system’s functionality, permissions, or scope of autonomous action is excessive depends on the task, risks, permission design, and operating environment.
Evidence
View evidence →
LLM06:2025 Excessive Agency
- Title
- LLM06:2025 Excessive Agency
- Author / organization
- OWASP
- Year
- 2025
- Source type / publication status
- Institutional security guidance; living/current
- Supported concepts / claim context
- Excessive Agency; functionality, permissions, autonomy and mitigations.
- Scope and limitations
- Security-framework terminology; not an ISO/JIS or universal terminology standard.
- Access / version note
- Living guidance; exact page URL not recorded.
- Recorded official URL / DOI / identifier
- As recorded:
NOT_RECORDED_IN_RESEARCH_INPUT (official OWASP page)(no URL inferred)
Agentic AI
- Title
- Agentic AI
- Author / organization
- NIST
- Year
- 2026
- Source type / publication status
- Government official web; current
- Supported concepts / claim context
- AI Agent; autonomy and goal-driven interaction.
- Scope and limitations
- A high-level official description, not a universally binding implementation standard.
- Access / version note
- Official page referenced in the research input; exact link not recorded.
- Recorded official URL / DOI / identifier
- As recorded:
NOT_RECORDED_IN_RESEARCH_INPUT (official NIST page)(no URL inferred)
Strengthening AI Agent Hijacking Evaluations
- Title
- Strengthening AI Agent Hijacking Evaluations
- Author / organization
- NIST CAISI
- Year
- 2025
- Source type / publication status
- Government primary evaluation / official blog; current
- Supported concepts / claim context
- Indirect Prompt Injection; agent hijacking.
- Scope and limitations
- AgentDojo simulated environments; findings are simulation-scoped.
- Access / version note
- Official page referenced; exact URL not recorded.
- Recorded official URL / DOI / identifier
- As recorded:
NOT_RECORDED_IN_RESEARCH_INPUT (official NIST page)(no URL inferred)