What is excessive agency?
Excessive agency is the condition in which an AI agent can do more than its task requires. OWASP lists it as LLM06:2025 and names three causes: excessive functionality, excessive permissions and excessive autonomy. It turns a model error or a successful prompt injection into a damaging action.
Also called over-permissioned agents, least privilege for AI agents. Last reviewed .
Why it matters
OWASP describes it as the vulnerability that lets damaging actions be performed in response to unexpected, ambiguous or manipulated model output. The agent does not need to be attacked for this to matter.
In July 2025, The Register reported that SaaStr founder Jason Lemkin said the Replit coding agent deleted his production database after he had told it not to change anything without permission. Replit responded by separating development and production databases, which is a limit on what the agent can reach.
Related terms
- Agent identity
- Agent identity is where the excess permissions live: the tokens and grants an agent holds.
- Prompt injection
- Prompt injection is the usual trigger that turns excess access into harm.
- Confused deputy
- A confused deputy is an agent whose broad access is used on behalf of someone with less.
- AI agent observability
- Observability shows which permissions an agent uses and which it never has.
How Guard0 handles it
Guard0 lists the tokens, grants and scopes each agent holds beside how often the agent has used each one. The person who answers for the agent can remove what it does not need, and the change stays on the record of that agent.
Sources
- LLM06:2025 Excessive Agency. OWASP Gen AI Security Project.
- Vibe coding service Replit deleted production database. The Register.
- OWASP Non-Human Identities Top 10 (2025). OWASP.