AAK-TAINT-006: Tool parameter flows to deserialization
- Severity
- MEDIUM
- Category
- Taint Analysis (
taint-analysis) - Scanner
agent_audit_kit/scanners/taint_analysis.py- Auto-fixable
- No.
- SARIF name
TaintDeserialization
What it detects
A @tool function parameter is passed to pickle.loads(), yaml.load(), or similar deserialization functions.
How to fix
Use safe deserialization (yaml.safe_load, json.loads) instead of unsafe alternatives.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-TAINT-006
Limitations
Only functions decorated @tool, @mcp.tool, @server.tool, @langchain.tools.tool, @crewai.tool or @autogen.tool, and a low-level @<server>.call_tool() handler, are analysed, so a FastMCP instance under another name (@app.tool()) is skipped. Reports pickle.load/loads, yaml.load/unsafe_load and marshal.loads when the argument is a parameter, an item read from one or, in a call_tool handler, a local holding one at most two assignments away. A local in a @tool function (data = blob) is not followed. Other loaders that can run code from the data are not sinks: torch.load, joblib.load and dill.loads are not reported.
References
- OWASP MCP Top 10: MCP05:2025 (Command Injection & Execution)
- OWASP Agentic Top 10: ASI05 (Unexpected Code Execution (RCE))
- Adversa AI MCP Top 25 (AAK-defined ids): ADV-INJECT-08 (Unsafe Deserialization)
- CSA AI Controls Matrix (AICM): AIS-07
All rules. This page is generated from the rule registry when the documentation is built; to change it, change the rule in agent_audit_kit/rules/builtin.py. To give the rule a hand-written write-up, add docs/rules/AAK-TAINT-006.md: the page then shows that text, with the registry reference appended under it.