AI
Atlas AI
JK
← Template library
OWASP LLM Top 10LLM06:2025highv1.0.0 · System

Agent Tool Restriction

Restricts which tools an agent can invoke based on user role and application context. Prevents agents from overstepping their authorised scope.

📘Clone & start observing

Creates a Guideline policy. Observation only — nothing is blocked until you promote to Strict.

Mode on clone: log
Defaults to template name. Customise to distinguish multiple instances of the same template.
Leave empty to apply broadly via the template's default data-classification / risk-tier filters.
Rationale

Agentic AI with broad tool access creates excessive agency risk. A customer-service agent should not be able to invoke financial-transaction tools.

Example violation
Customer Service agent attempting to call "transferFunds(account=ACC123, amount=10000)" tool
Triggers (1)
  • tool_callInspect every tool invocation against allowed scope
Detectors (1)
  • keyword_listtool-allowlist
    Compare tool name against per-role allowlist
Actions (3)
  • blockRefuse the tool call
  • logRecord attempted out-of-scope call
  • notifyAlert agent owner
Tunable parameters (3)
Allowed tools (per role)
basickeywords
Comma-separated list of tool names this agent may invoke.
Default: ["search_kb","fetch_user_profile","create_ticket"]
Block unknown tools
advancedboolean
Block any tool not on the allowlist. Disable only for development.
Default: true
Notification channel
basicchannel
Where to send out-of-scope alerts.
Default: "#agent-security"
Regulatory references
EU AI Act Art. 14
Template defaults (suggested target after promotion)
Suggested mode
block
Risk tiers
High-Risk
Data classifications
Departments

Cloned policies start in Guideline mode. Use the promotion wizard to flip to Strict once you trust the false-positive rate.