Recursive agent without iteration cap
ID |
recursive-agent-without-iteration-cap |
Severity |
high (catalogue Medium; platform severity model has no 'medium') |
Remediation Complexity |
trivial |
Remediation Risk |
low |
Remediation Effort |
low |
OWASP LLM |
LLM10:2025 — Unbounded Consumption |
OWASP ASI |
ASI08:2026 |
Family |
LLM10 — Unbounded Consumption |
Red-team vectors |
(none — LLM10 carries no red-team vector) |
Tags |
ai_security |
Description
An agent plan-act loop has no iteration / recursion / tool-call bound, so it can run forever — runaway cost, and the only static ASI08 coverage in the catalogue.
Framework-default aware: LangChain AgentExecutor defaults max_iterations=15 and LangGraph defaults recursion_limit=25, so a framework executor is bounded by default and is not flagged — unless the default is explicitly disabled (max_iterations=None). A raw while True loop driving an agent/LLM call has no framework default, so it fires unless an explicit cap is present.
Examples
agent = create_react_agent(llm, tools)
while True: (1)
response = agent.invoke(state)| 1 | Raw agent loop with no iteration cap — flagged. Also flagged: AgentExecutor(…, max_iterations=None). |
A framework executor with its default cap, or with an explicit max_iterations=10, produces no finding.