vLLM : DoS via unvalidated mm_processor_kwargs in multimodal image processors (Isaac, InternVL, H2OVL, etc.)

Summary API clients can supply arbitrary mmprocessorkwargs per request. vLLM forwards these values through the multimodal rendering and processing pipeline to Hugging Face and custom image/video processor constructors and calls. Resource-sensitive values such as image size, patch size, patch coun...

GHSA-GX7P-2J49-HFQ4 vLLM : DoS via unvalidated mm_processor_kwargs in multimodal image processors (Isaac, InternVL, H2OVL, etc.)

Summary API clients can supply arbitrary mmprocessorkwargs per request. vLLM forwards these values through the multimodal rendering and processing pipeline to Hugging Face and custom image/video processor constructors and calls. Resource-sensitive values such as image size, patch size, patch coun...

EUVD-2026-107006

vLLM : DoS via unvalidated mmprocessorkwargs in multimodal image processors Isaac, InternVL, H2OVL, etc...

vLLM: Forged multimodal UUID + P0/P1 processor-cache eviction drift: unauthenticated remote crash of the whole vLLM engine

Summary Multimodal content parts imageurl, videourl, inputaudio, ... accept a user-supplied uuid field that vLLM uses verbatim as the item's cache identity mmhash. The two frontend/engine processor caches P0 "sender" cache in the API process, P1 "receiver" cache in the EngineCore process are...

GHSA-P92P-RXJ5-7P2X vLLM: Forged multimodal UUID + P0/P1 processor-cache eviction drift: unauthenticated remote crash of the whole vLLM engine

Summary Multimodal content parts imageurl, videourl, inputaudio, ... accept a user-supplied uuid field that vLLM uses verbatim as the item's cache identity mmhash. The two frontend/engine processor caches P0 "sender" cache in the API process, P1 "receiver" cache in the EngineCore process are...

EUVD-2026-107001

vLLM: Forged multimodal UUID + P0/P1 processor-cache eviction drift: unauthenticated remote crash of the whole vLLM engine...

CVE-2024-11040

CVE-2024-11040 vllm-project vllm version 0.5.2.2 is vulnerable to Denial of Service attacks...

vLLM <= 0.23.0 - Anthropic Router Heap Address Information Leak

vLLM = 0.23.0 incompletely fixes CVE-2026-22778. The original fix added sanitizemessage to the OpenAI router but the Anthropic-compatible router /v1/messages echoes strexc directly. id: CVE-2026-54236 info: name: vLLM = 0.23.0 - Anthropic Router Heap Address Information Leak author: kenlacroix...

EXPLOIT-CVE-2026-22778

CVE-2026-22778 — vLLM RCE in video processing Vulnerable lab + proof of concept for CVE-2026-22778 CVSS 9.8, an unauthenticated remote code execution chain in vLLM's multimodal ingestion path. | ---|--- CVE| CVE-2026-22778 Advisory| GHSA-4r2x-xpjr-7cvv Affected| vLLM = 0.8.3, vLLM turned...

CVE-2026-114050

A flaw was found in vLLM. An unauthenticated remote attacker can cause a Denial of Service DoS by sending crafted multimodal requests with manipulated identifiers to desynchronize internal processor caches. This desynchronization causes an unhandled token mismatch error in the engine core,...

CVE-2026-114055

A flaw was found in vLLM. A remote attacker can cause a Denial of Service DoS by supplying manipulated multimodal processing parameters with excessively large image, frame, or patch values. Because the application processes these unvalidated parameters directly, it exhausts available memory and...

CVE-2026-114052

A flaw was found in vLLM. This vulnerability allows an unauthenticated remote attacker to cause a Denial of Service DoS via memory exhaustion. When serving MOSS-Audio models, the unauthenticated tokenize endpoint caches processor configurations in an unbounded cache. By submitting requests with...

CVE-2026-114051

A flaw was found in vllm. A remote attacker can cause a Denial of Service DoS by sending a specially crafted inference request containing an oversized minimum token parameter and an omitted maximum token parameter. Due to improper input validation, this request triggers an unhandled exception tha...

CVE-2026-114049

A flaw was found in vLLM. A remote attacker can compromise the integrity of responses generated for other users by submitting concurrent constrained generation requests. Due to improper state cleanup during batch reorganization, restriction rules defining permitted response tokens can persist in...

CVE-2026-114054

A flaw was found in vLLM. An authenticated remote attacker can cause shared cache key collisions by providing caller-controlled parameters, such as image metadata identifiers or cache salts. Successful exploitation allows requests to incorrectly reuse cached data from other operations, leading to...

CVE-2026-22807_Range

🛡️ CVE-2026-22807 Simulation: AI Supply Chain RCE Target Range ⚠️ Disclaimer This project is intended solely for security research and educational purposes , aiming to demonstrate the supply chain attack principle Remote Code Execution during AI model loading. Do not use the generated malicious...

Duplicate Advisory: vLLM 0.7.3 before 0.31.0 DoS via Unvalidated mm_processor_kwargs

Duplicate Advisory This advisory has been withdrawn because it is a duplicate of GHSA-gx7p-2j49-hfq4. This link is maintained to preserve external references. Original Description vLLM from 0.7.3 before 0.31.0 contains an uncontrolled resource consumption vulnerability that allows remote clients ...

Duplicate Advisory: vLLM vulnerable to remote denial of service through forged multimodal UUID cache desynchronization

Duplicate Advisory This advisory has been withdrawn because it is a duplicate of GHSA-p92p-rxj5-7p2x. This link is maintained to preserve external references. Original Description vLLM before 0.31.0 uses a client-supplied multimodal uuid verbatim as cache identity with no consistency check on cac...

Arbitrary Code Injection

Overview vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs Affected versions of this package are vulnerable to Arbitrary Code Injection via the /tokenize API endpoint, which does not enforce authentication even when the --api-key option is configured. An...

Expected Behavior Violation

Overview vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs Affected versions of this package are vulnerable to Expected Behavior Violation in the condense method of vllm/v1/worker/gpuinputbatch.py, where stale allowedtokenids mask entries are not cleared afte...