Browse / Direct Prompt Injection / CVE-2023-29374
LangChain vulnerable to code injection
- Class
- Direct Prompt Injection (filed by a model that read the advisory)
- Technique
- Code and command execution
- Also involves
- Improper Output Handling
- Affected
- langchain (PyPI)
- CVSS
- CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
- EPSS
- 39.7% chance of exploitation in 30 days
How it works, in short
Crafted user input steers the math chain's model into emitting Python that the chain then passes to exec unchecked. (Our summary, written by a model. The source text is below.)
Details
In LangChain through 0.0.131, the LLMMathChain chain allows prompt injection attacks that can execute arbitrary code via the Python exec() method.
References
- https://nvd.nist.gov/vuln/detail/CVE-2023-29374
- https://github.com/hwchase17/langchain/issues/1026
- https://github.com/hwchase17/langchain/issues/814
- https://github.com/hwchase17/langchain/pull/1119
- https://github.com/langchain-ai/langchain
- https://github.com/pypa/advisory-database/tree/main/vulns/langchain/PYSEC-2023-18.yaml
- https://twitter.com/rharang/status/1641899743608463365/photo/1
Source: GitHub Advisory Database, used under CC BY 4.0.