| Title | langchain-ai langchain 0.1.9 Remote Code Execution (RCE) |
|---|
| Description | A command execution vulnerability exists in the `langchain` project version `0.1.9`.
The code in line `106` of the `libs/community/langchain_community/retrievers/tfidf.py` file in this project have a function named `load_local`. There are two unsecured functions `joblib.load` and `pickle.load` in this function, and By constructing the malicious file, I generated the `tfidf_vectorizer.jobli`b and `tfidf_vectorizer.pkl` files in the PoC.pkl directory. They both contain strings that can be executed remotely by command., which, by not using a more secure way of loading and checking external files, resulted in being able to trigger a command execution that triggers an RCE via the commands contained in the malicious files. |
|---|
| Source | ⚠️ https://github.com/bayuncao/vul-cve-16 |
|---|
| User | bayuncao (UID 50143) |
|---|
| Submission | 02/28/2024 04:31 (2 years ago) |
|---|
| Moderation | 03/01/2024 07:41 (2 days later) |
|---|
| Status | Accepted |
|---|
| VulDB entry | 255372 [LangChain langchain_community 0.0.26 TFIDFRetriever tfidf.py load_local server-side request forgery] |
|---|
| Points | 20 |
|---|