OpenAI Agents SDK agent with tools but no guardrails
MediumAV-FW-006Human oversightFramework pack
What it finds
An agent built with the OpenAI Agents SDK has tools, but no input or output guardrails are defined.
Why it matters
Guardrails are the SDK's built-in way to stop off-topic, malicious, or unsafe requests before tools run. Without them, every request goes straight to the model and its tools.
What an attacker could do: Malicious or off-policy requests reach the agent's tools unchecked.
How to fix it
Add input_guardrails (block injection and off-topic requests) and output_guardrails (block sensitive data leaving), and require approval for consequential tools.
Pro shows the corrected code for your exact file, step by step, and re-scans to confirm the fix. See plans
Check details
- Finding code
- OPENAI_AGENTS_NO_GUARDRAILS
- How it detects
- Looks for a risky pattern, and stays quiet when a matching safeguard is present.
- Confidence
- Likely: strong signal, but review it in context.
- Scans
- Agent source using the named framework (LangChain, CrewAI, AutoGen, OpenAI Agents SDK, Claude Agent SDK, computer-use tools)
- OWASP LLM Top 10
- LLM01 - Prompt Injection, LLM08 - Excessive Agency
- NIST AI RMF
- MG-2.2, MS-2.3
- SOC 2
- CC7.2
Check your agent for this
Paste your agent code or MCP config. The free scan runs all 124 checks in about 10 seconds — no account needed.
Related checks
- Agent can move money with no amount limitAV-HUMAN-003 · Critical
- Missing human-in-the-loop gatesAV-HUMAN-001 · High
- Built-in safety controls turned offAV-HUMAN-002 · High
- Destructive tool with no confirmationAV-HUMAN-004 · High
- AI editor set to run every tool without askingAV-FW-013 · High
- MCP tools set to run without askingAV-MCP-104 · Medium