Vulnerability · Published May 10, 2024 · Assigned by GitHub_M
CVE-2024-34359llama-cpp-python vulnerable to Remote Code Execution by Server-Side Template Injection in Model Metadata
Critical 9.7
In plain English
A critical-severity security flaw in llama-cpp-python from abetlen. Proof-of-concept exploit code is public. The chance it is exploited in the next 30 days is estimated at 26.0% (higher than 98% of all known vulnerabilities). Fixed versions are available - upgrade the affected package.
Official description
llama-cpp-python is the Python bindings for llama.cpp. `llama-cpp-python` depends on class `Llama` in `llama.py` to load `.gguf` llama.cpp or Latency Machine Learning Models. The `__init__` constructor built in the `Llama` takes several parameters to configure the loading and running of the model. Other than `NUMA, LoRa settings`, `loading tokenizers,` and `hardware settings`, `__init__` also loads the `chat template` from targeted `.gguf` 's Metadata and furtherly parses it to `llama_chat_format.Jinja2ChatFormatter.to_chat_handler()` to construct the `self.chat_handler` for this model. Nevertheless, `Jinja2ChatFormatter` parse the `chat template` within the Metadate with sandbox-less `jinja2.Environment`, which is furthermore rendered in `__call__` to construct the `prompt` of interaction. This allows `jinja2` Server Side Template Injection which leads to remote code execution by a carefully constructed payload.
Affected packages & fixed versions
| Ecosystem | Package | Fixed in | Advisory |
|---|---|---|---|
| PyPI | llama-cpp-python | 0.2.72 | GHSA-56xg-wfcc-g829 |
| PyPI | llama-cpp-python | 0.2.72 | PYSEC-2026-392 |
Affected products (as reported)
| Vendor | Product | Versions |
|---|---|---|
| abetlen | llama-cpp-python | >= 0.2.30, <= 0.2.71 |
Weakness: CWE-76 · Vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:H