Skip to content
Heist/ feed
Back to feed
Critical

llama-cpp-python vulnerable to Remote Code Execution by Server-Side Template Injection in Model Metadata

Type: advisorySources: OSV

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.

Products
Not specified
Affected versions
Not specified
Fixed versions
Not specified
CVEs
CVE-2024-34359
Confirmed fixed CVEs
Not specified
Updated in this feed
6 Nov 2026, 00:58 UTC · Revision 1

Source evidence

llama-cpp-python vulnerable to Remote Code Execution by Server-Side Template Injection in Model Metadata

core-osv · Original source

Published 10 May 2024, 17:07 UTC

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.

osv:CVE-2024-34359
Complete source record
{
  "id": "3f626f90977426e71c48d81fececba99985fea6d7566caaae7385b2b3a90c4f6",
  "url": "https://osv.dev/vulnerability/CVE-2024-34359",
  "method": "structured",
  "payload": {
    "kind": "advisory",
    "title": "llama-cpp-python vulnerable to Remote Code Execution by Server-Side Template Injection in Model Metadata",
    "summary": "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.",
    "evidence": "osv:CVE-2024-34359",
    "products": [],
    "severity": "critical",
    "structured": {
      "id": "CVE-2024-34359",
      "aliases": [
        "CVE-2024-34359",
        "GHSA-56xg-wfcc-g829",
        "PYSEC-2026-392"
      ],
      "metrics": [
        {
          "type": "CVSS_V3",
          "score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:H"
        }
      ],
      "affected": [
        {
          "ranges": [
            {
              "repo": "https://github.com/abetlen/llama-cpp-python",
              "type": "GIT",
              "events": [
                {
                  "introduced": "656f3d896845d428f994dd205b28d8f524cbff67"
                },
                {
                  "fixed": "b454f40a9a1787b2b5659cd2cb00819d983185df"
                }
              ],
              "database_specific": {
                "extracted_events": [
                  {
                    "range": [
                      {
                        "introduced": "0.2.30"
                      },
                      {
                        "last_affected": "0.2.71"
                      }
                    ],
                    "source": "AFFECTED_FIELD",
                    "strategy": "StringRangeExpression"
                  },
                  {
                    "range": [
                      {
                        "fixed": "b454f40a9a1787b2b5659cd2cb00819d983185df"
                      }
                    ],
                    "source": "REFERENCES_COMMIT"
                  }
                ]
              }
            }
          ],
          "versions": [
            "v0.2.71-metal",
            "v0.2.71-cu124",
            "v0.2.71-cu123",
            "v0.2.71-cu122",
            "v0.2.71-cu121",
            "v0.2.71",
            "v0.2.70-metal",
            "v0.2.70-cu124",
            "v0.2.70-cu123",
            "v0.2.70-cu122",
            "v0.2.70-cu121",
            "v0.2.70",
            "v0.2.69-metal",
            "v0.2.69-cu124",
            "v0.2.69-cu123",
            "v0.2.69-cu122",
            "v0.2.69-cu121",
            "v0.2.69",
            "v0.2.68-metal",
            "v0.2.68-cu124",
            "v0.2.68-cu123",
            "v0.2.68-cu122",
            "v0.2.68-cu121",
            "v0.2.68",
            "v0.2.67-metal",
            "v0.2.67-cu124",
            "v0.2.67-cu123",
            "v0.2.67-cu122",
            "v0.2.67-cu121",
            "v0.2.67",
            "v0.2.66-metal",
            "v0.2.66-cu124",
            "v0.2.66-cu123",
            "v0.2.66-cu122",
            "v0.2.66-cu121",
            "v0.2.66",
            "v0.2.65-metal",
            "v0.2.65-cu123",
            "v0.2.65-cu122",
            "v0.2.65-cu121",
            "v0.2.65",
            "v0.2.64-metal",
            "v0.2.64-cu123",
            "v0.2.64-cu122",
            "v0.2.64-cu121",
            "v0.2.64",
            "v0.2.62-metal",
            "v0.2.62-cu123",
            "v0.2.62-cu122",
            "v0.2.62-cu121",
            "v0.2.62",
            "v0.2.61-metal",
            "v0.2.61-cu123",
            "v0.2.61-cu122",
            "v0.2.61-cu121",
            "v0.2.61",
            "v0.2.60-metal",
            "v0.2.60-cu123",
            "v0.2.60-cu122",
            "v0.2.60-cu121",
            "v0.2.60",
            "v0.2.59-metal",
            "v0.2.59-cu123",
            "v0.2.59-cu122",
            "v0.2.59-cu121",
            "v0.2.59",
            "v0.2.58",
            "v0.2.57",
            "v0.2.56",
            "v0.2.55",
            "v0.2.54",
            "v0.2.53",
            "v0.2.52",
            "v0.2.51",
            "v0.2.50",
            "v0.2.49",
            "v0.2.48",
            "v0.2.47",
            "v0.2.46",
            "v0.2.45",
            "v0.2.44",
            "v0.2.43",
            "v0.2.42",
            "v0.2.41",
            "v0.2.40",
            "v0.2.39",
            "v0.2.38",
            "v0.2.37",
            "v0.2.36",
            "v0.2.35",
            "v0.2.34",
            "v0.2.33",
            "v0.2.32",
            "v0.2.31",
            "v0.2.30"
          ],
          "database_specific": {
            "source": "https://storage.googleapis.com/cve-osv-conversion/osv-output/CVE-2024-34359.json"
          }
        }
      ],
      "modified": "2026-10-28T02:30:45.596087141Z",
      "provider": "osv",
      "references": [
        "https://github.com/CVEProject/cvelistV5/tree/main/cves/2024/34xxx/CVE-2024-34359.json",
        "https://github.com/abetlen/llama-cpp-python/security/advisories/GHSA-56xg-wfcc-g829",
        "https://nvd.nist.gov/vuln/detail/CVE-2024-34359",
        "https://github.com/abetlen/llama-cpp-python/commit/b454f40a9a1787b2b5659cd2cb00819d983185df"
      ],
      "provider_metadata": {
        "cwe_ids": [
          "CWE-76"
        ],
        "cna_assigner": "GitHub_M",
        "osv_generated_from": "https://github.com/CVEProject/cvelistV5/tree/main/cves/2024/34xxx/CVE-2024-34359.json"
      }
    },
    "related_cves": [
      {
        "value": "CVE-2024-34359",
        "evidence": "/id"
      }
    ],
    "security_relevant": true,
    "severity_evidence": "/severity/0/score"
  },
  "source_id": "core-osv",
  "document_id": 513668,
  "published_at": "2024-05-10T17:07:18.85+00:00"
}