GHSA-Q8GQ-377P-JQ3R

Vulnerability from github – Published: 2026-06-16 17:34 – Updated: 2026-06-16 17:34
VLAI
Summary
vLLM: Security Check Bypass via assert Statement in Activation Function Loading Allows Arbitrary Code Execution
Details

Summary

An assert-based security check in vLLM's activation function loading allows any unauthenticated attacker to achieve arbitrary code execution on the server by publishing a malicious HuggingFace model, when vLLM runs in Python optimized mode (python -O or PYTHONOPTIMIZE=1).

Details

vLLM uses an assert statement at vllm/model_executor/layers/pooler/activations.py:48 as its sole security control to restrict which activation functions can be loaded from a HuggingFace model's config.json:

# vllm/model_executor/layers/pooler/activations.py:35-53
function_name: str | None = None
if (
    hasattr(config, "sentence_transformers")
    and "activation_fn" in config.sentence_transformers
):
    function_name = config.sentence_transformers["activation_fn"]
elif (
    hasattr(config, "sbert_ce_default_activation_function")
    and config.sbert_ce_default_activation_function is not None
):
    function_name = config.sbert_ce_default_activation_function

if function_name is not None:
    assert function_name.startswith("torch.nn.modules."), (
        "Loading of activation functions is restricted to "
        "torch.nn.modules for security reasons"
    )
    fn = resolve_obj_by_qualname(function_name)()

Python's assert statements are stripped at compile time when running in optimized mode (python -O or PYTHONOPTIMIZE=1). When the assert is absent, the attacker-controlled function_name from the model's config.json is passed directly to resolve_obj_by_qualname() — an unrestricted import gadget:

def resolve_obj_by_qualname(qualname: str) -> Any:
    module_name, obj_name = qualname.rsplit(".", 1)
    module = importlib.import_module(module_name)
    return getattr(module, obj_name)

This is the same vulnerability class as CVE-2017-1000433 (pysaml2 assert-based auth bypass), flagged by Bandit B101 and Ruff S101, and the reason Django proactively replaced all assert-based security checks (ticket #32508).

Attacker-controlled input sources: - config.sentence_transformers["activation_fn"] (line 40) - config.sbert_ce_default_activation_function (line 45)

Affected call sites — get_act_fn() is called via resolve_classifier_act_fn() from: - vllm/model_executor/layers/pooler/seqwise/poolers.py:122 — SequencePooler - vllm/model_executor/layers/pooler/tokwise/poolers.py:130 — TokenPooler

Broader systemic risk: resolve_obj_by_qualname is called from ~20 locations across the codebase with no validation of its own. Any future caller feeding user-controlled input to it without validation creates the same vulnerability class.

Suggested fix: Replace the assert with an explicit conditional raise:

if not function_name.startswith("torch.nn.modules."):
    raise ValueError(
        "Loading of activation functions is restricted to "
        "torch.nn.modules for security reasons"
    )

Impact

Arbitrary code execution. A malicious model author publishes a HuggingFace model with a crafted config.json. When a victim loads this model with vLLM running under python -O or PYTHONOPTIMIZE=1, arbitrary code executes during model initialization with the privileges of the vLLM process.

The attack requires: 1. Victim loads a malicious model from HuggingFace (user interaction) 2. vLLM runs under python -O or PYTHONOPTIMIZE=1 (documented in production use) 3. Model uses a cross-encoder architecture (e.g. BERT or RoBERTa with sequence classification)

Coordinated disclosure note: This vulnerability was also reported via huntr.com on April 2, 2026 (https://huntr.com/bounties/dcb05b04-e625-41e7-adbc-bbae0cc2d64c). A GitHub Security Advisory was also filed because it is vLLM's stated preferred disclosure channel per SECURITY.md.

Fix

A fix for this was introduced in this commit: https://github.com/vllm-project/vllm/commit/b3c7ffcab82c2439726f8cb213800f6f38c023d3

Severity
7.5 (High)
CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:U/C:H/I:H/A:H
Show details on source website
To clipboard 

{
  "affected": [
    {
      "package": {
        "ecosystem": "PyPI",
        "name": "vllm"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0"
            },
            {
              "fixed": "0.22.0"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-41523"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-617",
      "CWE-94"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-06-16T17:34:49Z",
    "nvd_published_at": null,
    "severity": "HIGH"
  },
  "details": "### Summary\n\nAn `assert`-based security check in vLLM\u0027s activation function loading allows any unauthenticated attacker to achieve arbitrary code execution on the server by publishing a malicious HuggingFace model, when vLLM runs in Python optimized mode (`python -O` or `PYTHONOPTIMIZE=1`).\n\n### Details\n\nvLLM uses an `assert` statement at [`vllm/model_executor/layers/pooler/activations.py:48`](https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/layers/pooler/activations.py#L48) as its sole security control to restrict which activation functions can be loaded from a HuggingFace model\u0027s `config.json`:\n\n```python\n# vllm/model_executor/layers/pooler/activations.py:35-53\nfunction_name: str | None = None\nif (\n    hasattr(config, \"sentence_transformers\")\n    and \"activation_fn\" in config.sentence_transformers\n):\n    function_name = config.sentence_transformers[\"activation_fn\"]\nelif (\n    hasattr(config, \"sbert_ce_default_activation_function\")\n    and config.sbert_ce_default_activation_function is not None\n):\n    function_name = config.sbert_ce_default_activation_function\n\nif function_name is not None:\n    assert function_name.startswith(\"torch.nn.modules.\"), (\n        \"Loading of activation functions is restricted to \"\n        \"torch.nn.modules for security reasons\"\n    )\n    fn = resolve_obj_by_qualname(function_name)()\n```\n\nPython\u0027s `assert` statements are stripped at compile time when running in optimized mode (`python -O` or `PYTHONOPTIMIZE=1`). When the assert is absent, the attacker-controlled `function_name` from the model\u0027s `config.json` is passed directly to [`resolve_obj_by_qualname()`](https://github.com/vllm-project/vllm/blob/main/vllm/utils/import_utils.py#L106) \u2014 an unrestricted import gadget:\n\n```python\ndef resolve_obj_by_qualname(qualname: str) -\u003e Any:\n    module_name, obj_name = qualname.rsplit(\".\", 1)\n    module = importlib.import_module(module_name)\n    return getattr(module, obj_name)\n```\n\nThis is the same vulnerability class as **CVE-2017-1000433** (pysaml2 assert-based auth bypass), flagged by Bandit B101 and Ruff S101, and the reason Django proactively replaced all assert-based security checks (ticket #32508).\n\n**Attacker-controlled input sources:**\n- `config.sentence_transformers[\"activation_fn\"]` (line 40)\n- `config.sbert_ce_default_activation_function` (line 45)\n\n**Affected call sites** \u2014 `get_act_fn()` is called via `resolve_classifier_act_fn()` from:\n- `vllm/model_executor/layers/pooler/seqwise/poolers.py:122` \u2014 SequencePooler\n- `vllm/model_executor/layers/pooler/tokwise/poolers.py:130` \u2014 TokenPooler\n\n**Broader systemic risk:** `resolve_obj_by_qualname` is called from ~20 locations across the codebase with no validation of its own. Any future caller feeding user-controlled input to it without validation creates the same vulnerability class.\n\n**Suggested fix:** Replace the `assert` with an explicit conditional raise:\n\n```python\nif not function_name.startswith(\"torch.nn.modules.\"):\n    raise ValueError(\n        \"Loading of activation functions is restricted to \"\n        \"torch.nn.modules for security reasons\"\n    )\n```\n\n### Impact\n\n**Arbitrary code execution.** A malicious model author publishes a HuggingFace model with a crafted `config.json`. When a victim loads this model with vLLM running under `python -O` or `PYTHONOPTIMIZE=1`, arbitrary code executes during model initialization with the privileges of the vLLM process.\n\nThe attack requires:\n1. Victim loads a malicious model from HuggingFace (user interaction)\n2. vLLM runs under `python -O` or `PYTHONOPTIMIZE=1` (documented in production use)\n3. Model uses a cross-encoder architecture (e.g. BERT or RoBERTa with sequence classification)\n\n**Coordinated disclosure note:** This vulnerability was also reported via huntr.com on April 2, 2026 (https://huntr.com/bounties/dcb05b04-e625-41e7-adbc-bbae0cc2d64c). A GitHub Security Advisory was also filed because it is vLLM\u0027s stated preferred disclosure channel per SECURITY.md.\n\n### Fix\n\nA fix for this was introduced in this commit: https://github.com/vllm-project/vllm/commit/b3c7ffcab82c2439726f8cb213800f6f38c023d3",
  "id": "GHSA-q8gq-377p-jq3r",
  "modified": "2026-06-16T17:34:49Z",
  "published": "2026-06-16T17:34:49Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-q8gq-377p-jq3r"
    },
    {
      "type": "WEB",
      "url": "https://github.com/vllm-project/vllm/commit/b3c7ffcab82c2439726f8cb213800f6f38c023d3"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/vllm-project/vllm"
    },
    {
      "type": "WEB",
      "url": "https://huntr.com/bounties/dcb05b04-e625-41e7-adbc-bbae0cc2d64c"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:U/C:H/I:H/A:H",
      "type": "CVSS_V3"
    }
  ],
  "summary": "vLLM: Security Check Bypass via assert Statement in Activation Function Loading Allows Arbitrary Code Execution"
}


Log in or create an account to share your comment.




Tags
Taxonomy of the tags.





Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.

Sightings

Author Source Type Date Other

Nomenclature

  • Seen: The vulnerability was mentioned, discussed, or observed by the user.
  • Confirmed: The vulnerability has been validated from an analyst's perspective.
  • Published Proof of Concept: A public proof of concept is available for this vulnerability.
  • Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
  • Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
  • Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
  • Not confirmed: The user expressed doubt about the validity of the vulnerability.
  • Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.


Detection rules are retrieved from Rulezet.