Excessive agency
Japanese 過剰なエージェンシー(AIに任せすぎるリスク)
Excessive agency is a risk where an AI system has more functions, permissions or freedom than it needs, so unexpected or manipulated output can lead to harmful actions. It is LLM06 in the OWASP list of AI risks.
Excessive agency is the risk that an AI system has more functions and permissions than its job needs, and can act on its own without a person checking. If the AI misunderstands or is tricked into odd output, a lot of power turns that mistake into real damage. It is like handing a new employee every room key and the company stamp on day one.
For example, if an AI connected to read and summarize your email can also send email, one bad message could make it forward private content to an outsider. Limiting it to read-only access and having a person approve every send keeps the damage small.
Source: LLM06:2025 Excessive Agency · LLM01:2025 Prompt Injection (checked 2026-10-26)