CVE-2026-44222
published 2026-05-12CVE-2026-44222: vLLM is an inference and serving engine for large language models (LLMs). From 0.6.1 to before 0.20.0, there is a a Token Injection vulnerability in vLLM’s…
PriorityP345high7.5CVSS 3.1
AVNACLPRNUINSUCNINAH
EPSS
0.41%
34.5th percentile
vLLM is an inference and serving engine for large language models (LLMs). From 0.6.1 to before 0.20.0, there is a 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. This vulnerability is fixed in 0.20.0.
Affected
22 ranges
| Vendor | Product | Version range | Fixed in |
|---|---|---|---|
| rhaiis | vllm-cpu-rhel9 | — | — |
| rhaiis | vllm-cuda-rhel9 | — | — |
| rhaiis | vllm-neuron-rhel9 | — | — |
| rhaiis | vllm-rocm-rhel9 | — | — |
| rhaiis | vllm-spyre-rhel9 | — | — |
| rhaiis | vllm-tpu-rhel9 | — | — |
| rhelai3 | bootc-aws-cuda-rhel9 | — | — |
| rhelai3 | bootc-azure-cuda-rhel9 | — | — |
| rhelai3 | bootc-azure-rocm-rhel9 | — | — |
| rhelai3 | bootc-cuda-rhel9 | — | — |
| rhelai3 | bootc-gcp-cuda-rhel9 | — | — |
| rhelai3 | bootc-rocm-rhel9 | — | — |
| rhoai | odh-kserve-agent-rhel9 | — | — |
| rhoai | odh-kserve-controller-rhel9 | — | — |
| rhoai | odh-kserve-router-rhel9 | — | — |
| rhoai | odh-kserve-storage-initializer-rhel9 | — | — |
| rhoai | odh-vllm-cuda-rhel9 | — | — |
| rhoai | odh-vllm-gaudi-rhel9 | — | — |
| rhoai | odh-vllm-rocm-rhel9 | — | — |
| vllm-project | vllm | — | — |
| vllm | vllm | >= 0.6.1 < 0.20.0 | 0.20.0 |
| vllm | vllm | >= 0.6.1 < 0.20.0 | 0.20.0 |
CVSS provenance
nvdv3.17.5HIGHCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
vendor_redhat7.5HIGH
Stop checking back — get the weekly exploitation signal.
Every Monday: what got weaponized or added to CISA KEV in the last seven days — each CVE cross-linked to its PoC, Nuclei template, and detection rule. Free, one email a week, unsubscribe in one click.
Red Hat
vllm: vLLM: Denial of Service via malformed multimodal input or token injection
vendor_redhat·2026-05-12·CVSS 7.5
CVE-2026-44222 [HIGH] CWE-1285 vllm: vLLM: Denial of Service via malformed multimodal input or token injection
vllm: vLLM: Denial of Service via malformed multimodal input or token injection
vLLM is an inference and serving engine for large language models (LLMs). From 0.6.1 to before 0.20.0, there is a 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. This vulnerability is fixed in 0.20.0.
A flaw was found in vLLM, an inference and serving engine for large language models (LLMs). This vulnerability
VulDB
vllm-project vllm up to 0.19.x Placeholder image_grid_thw/video_grid_thw array index (ID 32656)
vuldb·2026-05-12
CVE-2026-44222 [LOW] vllm-project vllm up to 0.19.x Placeholder image_grid_thw/video_grid_thw array index (ID 32656)
A vulnerability was found in vllm-project vllm up to 0.19.x. It has been declared as problematic. This affects the function image_grid_thw/video_grid_thw of the component Placeholder Handler. Such manipulation leads to improper validation of array index.
This vulnerability is traded as CVE-2026-44222. The attack may be launched remotely. There is no exploit available.
It is recommended to upgrade the affected component.
GHSA
vLLM Vulnerable to Remote DoS via Special-Token Placeholders
ghsa·2026-05-05
CVE-2026-44222 [MEDIUM] CWE-129 vLLM Vulnerable to Remote DoS via Special-Token Placeholders
vLLM Vulnerable to Remote DoS via Special-Token Placeholders
## 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_
No detection rules found.
No public exploits indexed.
2026-05-12
Published