DATA · LIVING REFERENCE · T-DATA-CVE
AI/ML CVE Severity Tracker
Security vulnerabilities affecting the AI software supply chain — the libraries and serving infrastructure that modern LLM applications depend on. We query the NIST National Vulnerability Database for CVEs mentioning RAG stacks, model loaders, and inference servers — langchain, llama-index, transformers, PyTorch, TensorFlow, vLLM, Ollama, Hugging Face, Gradio, Streamlit, ComfyUI, Triton, and ONNX — over the last 90 days. Insecure model deserialization, server-side request forgery, and authentication bypasses are recurring themes: AI dependencies are now a first-class attack surface.
| CVE | CVSS | Severity | Affected | Summary | Published |
|---|---|---|---|---|---|
| CVE-2026-75128 | 9.8 | CRITICAL | langchain | Langchain-Chatchat through 0.3.1.3 exposes a POST /tools/call API endpoint that invokes any registered tool by name without authentication, leaving the always-registered shell tool callable. | 2026-10-20 |
| CVE-2026-51884 | 9.8 | CRITICAL | langchain | The /knowledge_base/upload_temp_docs temporary document upload endpoint in Langchain Chatchat 0.3.1 is vulnerable to path traversal. | 2026-10-01 |
| CVE-2026-51883 | 9.1 | CRITICAL | langchain | The knowledge base creation and document upload interfaces in Langchain-Chatchat 0.3.0;0.3.1 is vulnerable to path traversal. | 2026-10-01 |
| CVE-2026-51882 | 9.1 | CRITICAL | langchain | The OpenAI-compatible file upload endpoint `/v1/files` in Langchain-Chatchat 0.3.0 is vulnerable to path traversal. | 2026-10-01 |
| CVE-2026-61732 | 10.0 | CRITICAL | ollama | Decepticon is an autonomous hacking agent for red teams. | 2026-09-24 |
| CVE-2025-66455 | 9.8 | CRITICAL | pytorch | LMDeploy is a toolkit for compressing, deploying, and serving large language models. | 2026-09-18 |
| CVE-2026-47627 | 9.8 | CRITICAL | triton | NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker could cause path traversal. | 2026-08-18 |
| CVE-2026-126397 | 7.5 | HIGH | vllm | vLLM from 0.10.1 before 0.31.0 forwards a request-controlled mm_processor_kwargs.max_patches value to the Hugging Face processor for Cohere2 Vision models without a server-side upper bound, causing quadratic CPU work during multimodal prepr | 2026-11-13 |
| CVE-2026-123974 | 7.8 | HIGH | pytorch | In all versions of PyTorch's TorchServe, a predictable extraction directory in /tmp lets a local user swap the model handler and run code as the TorchServe service account leading to local privilege escalation. | 2026-11-06 |
| CVE-2026-123019 | 7.3 | HIGH | huggingface | A weakness has been identified in eosphoros-ai DB-GPT up to 0.8.2. | 2026-11-05 |
| CVE-2026-119978 | 8.5 | HIGH | langchain | Flowise through 3.1.4 contains a sandbox escape vulnerability that allows authenticated users to execute OS commands by abusing allowlisted langchain and mammoth modules when ALLOW_BUILTIN_DEP=true. | 2026-10-30 |
| CVE-2026-119408 | 8.8 | HIGH | langchain | n8n before 2.42.6, 2.43.x before 2.43.3, and 3.x before 3.0.1 contains a remote code execution vulnerability via deprecated Function, Function Item, and LangChain Code nodes running outside task runner isolation. | 2026-10-30 |
| CVE-2026-119120 | 7.5 | HIGH | pytorch, huggingface | Crawl4AI from 0.9.0 before 0.9.5 fails to filter the CosineStrategy model_name field in untrusted Docker API request bodies, allowing authenticated clients to choose which HuggingFace model the server loads. | 2026-10-29 |
| CVE-2026-59158 | 7.5 | HIGH | ollama | Nuxt Ollama integrates the official Ollama JavaScript library into Nuxt applications. | 2026-10-27 |
| CVE-2026-115924 | 8.6 | HIGH | onnx | Rembg is a tool to remove images background. | 2026-10-26 |
| CVE-2026-115915 | 8.1 | HIGH | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-26 |
| CVE-2026-115912 | 7.5 | HIGH | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-26 |
| CVE-2026-116180 | 7.7 | HIGH | vllm | vLLM from 0.8.5 before 0.31.0 has a denial of service flaw. | 2026-10-25 |
| CVE-2026-116179 | 7.5 | HIGH | vllm | vLLM from 0.18.0 before 0.31.0 contains a resource exhaustion vulnerability that allows API clients to bypass the video/jpeg frame-count cap by overriding num_frames through media_io_kwargs. | 2026-10-25 |
| CVE-2026-116177 | 7.5 | HIGH | vllm | vLLM from 0.10.2 before 0.25.0 contains an uncontrolled resource consumption vulnerability that allows unauthenticated attackers to exhaust memory because the Nemotron-VL processors globally set Image.MAX_IMAGE_PIXELS to None. | 2026-10-25 |
| CVE-2026-116176 | 8.2 | HIGH | vllm | vLLM from 0.9.2 before 0.24.0 contains a server-side request forgery vulnerability that allows API callers to force outbound connections by embedding a decode address in the X-Request-Id header. | 2026-10-25 |
| CVE-2026-114055 | 7.5 | HIGH | vllm | vLLM from 0.7.3 before 0.31.0 contains an uncontrolled resource consumption vulnerability that allows remote clients to control multimodal processor parameters by supplying per-request mm_processor_kwargs. | 2026-10-21 |
| CVE-2026-114053 | 8.1 | HIGH | transformers, vllm | vLLM from 0.7.3 before 0.31.0 forwards the request-supplied mm_processor_kwargs.code_revision to Transformers AutoProcessor.from_pretrained, letting clients select which revision of a model's dynamic processor code is imported when trust_re | 2026-10-21 |
| CVE-2026-114052 | 7.5 | HIGH | vllm | vLLM after 0.24.0 and before 0.31.0 contains a memory exhaustion vulnerability in MOSS-Audio model serving that caches a processor for every distinct request-controlled mm_processor_kwargs value in an unbounded dictionary. | 2026-10-21 |
| CVE-2026-114051 | 7.5 | HIGH | vllm | vLLM before 0.31.0 skips the min_tokens <= max_tokens check for /inference/v1/generate requests with a null max_tokens, so the EngineCore input thread later dies on an uncaught VLLMValidationError while /health still returns 200. | 2026-10-21 |
| CVE-2026-114050 | 7.5 | HIGH | vllm | vLLM before 0.31.0 uses a client-supplied multimodal uuid verbatim as cache identity with no consistency check on cache hits, allowing engine crashes. | 2026-10-21 |
| CVE-2026-110719 | 7.2 | HIGH | ollama | Portkey AI Gateway 1.14.0 through 1.15.2 contains a server-side request forgery vulnerability that allows unauthenticated attackers to reach loopback and Docker host services because isValidCustomHost trusts localhost-style hosts when TRUST | 2026-10-14 |
| CVE-2026-108836 | 7.3 | HIGH | ollama | A vulnerability has been found in Ollama up to 0.34.1. | 2026-10-12 |
| CVE-2026-102697 | 7.8 | HIGH | ollama | Ollama versions 0.14.0 before 0.31.2 contain an incorrect authorization vulnerability in the experimental agent mode Bash tool approval mechanism that fails to properly parse shell syntax. | 2026-09-29 |
| CVE-2026-94627 | 7.5 | HIGH | vllm | vLLM Mooncake connector through 0.29.0 fails to properly manage GPU KV cache block ownership when concurrent child requests share a single transfer ID in prefill/decode disaggregated deployments. | 2026-09-21 |
| CVE-2026-94626 | 7.5 | HIGH | vllm | vLLM through 0.29.0 fails to validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion endpoints, allowing attackers to allocate unbounded memory. | 2026-09-21 |
| CVE-2026-94624 | 7.5 | HIGH | vllm | vLLM through 0.29.0 contains a denial of service vulnerability in P2P KV offloading when OffloadingConnector is configured with TieringOffloadingSpec and a peer-to-peer secondary tier. | 2026-09-21 |
| CVE-2026-94623 | 7.5 | HIGH | vllm | vLLM through 0.29.0 contains a denial of service vulnerability in the NIXL connector's prefix caching implementation that fails to properly validate block counts across multi-prompt completion requests in prefill/decode disaggregated deploy | 2026-09-21 |
| CVE-2026-94622 | 7.5 | HIGH | vllm | vLLM versions through 0.29.0 contain a denial of service vulnerability in the NIXL connector's metadata handling for prefill/decode disaggregated deployments. | 2026-09-21 |
| CVE-2026-33625 | 8.8 | HIGH | huggingface | LMDeploy is a toolkit for compressing, deploying, and serving large language models. | 2026-09-18 |
| CVE-2026-93592 | 7.5 | HIGH | vllm | vLLM versions before 0.28.0 fail to validate the lower bound of token IDs in the /v1/embeddings and /pooling endpoints, allowing unauthenticated attackers to crash the engine by submitting negative token IDs. | 2026-09-18 |
| CVE-2026-93436 | 7.5 | HIGH | vllm | vLLM through 0.29.0 fails to properly clean up decode-side metadata for rejected inference requests in prefill/decode disaggregated deployments. | 2026-09-17 |
| CVE-2026-92816 | 7.8 | HIGH | comfyui | ComfyUI before 0.30.0 fails to sanitize folder_name input in dataset save nodes, allowing attackers to write files to arbitrary paths outside the output directory. | 2026-09-16 |
| CVE-2026-55253 | 7.7 | HIGH | langchain | LangChain MongoDB provides integrations between MongoDB, Atlas, LangChain, and LangGraph. | 2026-09-14 |
| CVE-2026-90553 | 7.8 | HIGH | vllm | vLLM before 0.28.0 contains a remote code execution vulnerability in the LlavaOnevision2 processor loader that ignores the trust_remote_code parameter when loading remote processor classes. | 2026-09-12 |
| CVE-2026-47625 | 7.5 | HIGH | triton | NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker could abuse missing authorization. | 2026-09-08 |
| CVE-2026-16497 | 7.5 | HIGH | triton | NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker could cause excessive iteration. | 2026-09-08 |
| CVE-2026-85180 | 7.5 | HIGH | ollama | Ollama fails to validate redirect destinations when pulling tensor-layer models, allowing unauthenticated attackers to redirect blob downloads to arbitrary hosts. | 2026-09-03 |
| CVE-2026-80047 | 7.8 | HIGH | transformers | A vulnerability in Hugging Face Transformers (versions 4.57.0 to 5.16.1) allows remote Python files to be written to local disk without user consent when using GenerativePreTrainedModel.load_custom_generate(). | 2026-09-01 |
| CVE-2026-82288 | 7.5 | HIGH | gradio | Stable Diffusion WebUI through 1.10.1 contains a credential disclosure vulnerability in the /sdapi/v1/cmd-flags endpoint that returns parsed command-line arguments including gradio_auth and api_auth values in cleartext. | 2026-08-28 |
| CVE-2026-82275 | 7.5 | HIGH | gradio | Qwen-Agent through 0.0.34 contains a path traversal vulnerability in the document parser that fails to restrict file access to intended directories. | 2026-08-28 |
| CVE-2026-82268 | 7.5 | HIGH | gradio | Qwen-Agent through 0.0.34 contains a server-side request forgery vulnerability in the document parsing path that treats caller-supplied paths as URLs without scheme restriction or host validation. | 2026-08-28 |
| CVE-2026-37237 | 7.5 | HIGH | vllm | vLLM up to and including 0.17.0 allows remote attackers to cause a Denial of Service via memory exhaustion. | 2026-08-28 |
| CVE-2026-47852 | 7.5 | HIGH | onnx | A local attacker on a multi-user host can pre-create the deterministic cache path and plant a malicious ONNX model file. | 2026-08-27 |
| CVE-2026-58474 | 8.8 | HIGH | huggingface | whichllm before 0.5.16 contains a code injection vulnerability in the run and snippet commands that allows a remote attacker who controls a HuggingFace repository to achieve arbitrary code execution by crafting a malicious GGUF filename con | 2026-08-26 |
| CVE-2026-79784 | 8.8 | HIGH | pytorch | Vocos instantiates a class named by a configuration file without restricting which class may be named. | 2026-08-25 |
| CVE-2026-76841 | 8.8 | HIGH | pytorch | Xinference loads models with Hugging Face remote code execution unconditionally enabled, and before version 2.12.0 exposes no setting to disable it. | 2026-08-24 |
| CVE-2026-49114 | 7.1 | HIGH | onnx | In ONNX before 1.21.0, the 'save_external_data' function builds the external-data file path from the model's external_data location field and opens it for writing without 'O_NOFOLLOW/O_EXCL', after a non-atomic 'os.path.isfile()' check. | 2026-08-21 |
| CVE-2026-72848 | 8.6 | HIGH | langchain | SitemapLoader.parse_sitemap in langchain_community/document_loaders/sitemap.py applies the documented restrict_to_same_domain control only to leaf url entries. | 2026-08-20 |
| CVE-2026-15679 | 7.8 | HIGH | pytorch | Hugging Face PyTorch Image Models checkpoint Deserialization of Untrusted Data Remote Code Execution Vulnerability. | 2026-08-20 |
| CVE-2026-47629 | 7.5 | HIGH | triton | NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker could cause improper input validation. | 2026-08-18 |
| CVE-2026-47628 | 7.5 | HIGH | triton | NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker could cause an allocation of resources without limits. | 2026-08-18 |
| CVE-2026-129356 | 5.0 | MEDIUM | vllm | A vulnerability was found in vllm-project vllm 0.30.0/0.31.0. | 2026-11-15 |
| CVE-2026-126399 | 6.5 | MEDIUM | vllm | vLLM from 0.19.0 before 0.31.0 contains a resource exhaustion vulnerability that allows inference clients to bypass the server frame limit by setting media_io_kwargs.video.num_frames in chat completion requests. | 2026-11-13 |
| CVE-2026-126398 | 5.9 | MEDIUM | vllm | vLLM from 0.28.0 before 0.30.0 contains a reachable assertion vulnerability in GPUModelRunner._init_xdrope_positions that allows remote attackers to crash the engine by submitting embeddings-only requests. | 2026-11-13 |
| CVE-2026-126396 | 5.3 | MEDIUM | vllm | vLLM from 0.24.0 before 0.31.0 contains a resource exhaustion vulnerability in the MiniMax M3 VL video processor, which accepts request-level min_pixels without range or min_pixels <= max_pixels validation. | 2026-11-13 |
| CVE-2026-126395 | 5.3 | MEDIUM | vllm | vLLM from 0.14.0rc1 before 0.31.0 contains an information disclosure vulnerability in sanitize_message() whose lowercase-only regex fails to scrub uppercase PIL image addresses. | 2026-11-13 |
| CVE-2026-126394 | 5.3 | MEDIUM | vllm | vLLM 0.30.0 before 0.31.0 contains a resource exhaustion vulnerability in OpenCVDynamicOpenPanguVideoBackend, which uses request-controlled num_frames without an effective ceiling. | 2026-11-13 |
| CVE-2026-125386 | 5.3 | MEDIUM | langchain | langchain4j-aideepin through 3.32.0 contains an insecure direct object reference vulnerability in CharacterMessageController that allows authenticated users to access other users' message references due to missing owner checks. | 2026-11-10 |
| CVE-2026-124588 | 6.3 | MEDIUM | huggingface | A security flaw has been discovered in huggingface finetrainers 0.1.0/0.2.0. | 2026-11-09 |
| CVE-2026-122160 | 6.3 | MEDIUM | huggingface | A security vulnerability has been detected in myshell-ai AIlice up to 0.4.0. | 2026-11-05 |
| CVE-2026-122159 | 6.3 | MEDIUM | onnx | A weakness has been identified in Edge0-AI Audio8_TTS up to 1d64afcc2da56222b41ddcf53de7b8bc54be1fb6. | 2026-11-05 |
| CVE-2026-121938 | 5.3 | MEDIUM | langchain | langchain4j-aideepin before 3.31.0 contains an insecure direct object reference vulnerability that allows authenticated users to read other users' character-chat conversations by supplying their character uuid. | 2026-11-04 |
| CVE-2026-115922 | 6.2 | MEDIUM | onnx | Open Neural Network Exchange (ONNX) is an open standard for machine learning interoperability. | 2026-10-26 |
| CVE-2026-115921 | 6.2 | MEDIUM | onnx | Open Neural Network Exchange (ONNX) is an open standard for machine learning interoperability. | 2026-10-26 |
| CVE-2026-115920 | 6.2 | MEDIUM | onnx | Open Neural Network Exchange (ONNX) is an open standard for machine learning interoperability. | 2026-10-26 |
| CVE-2026-115919 | 6.2 | MEDIUM | onnx | Open Neural Network Exchange (ONNX) is an open standard for machine learning interoperability. | 2026-10-26 |
| CVE-2026-115918 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-26 |
| CVE-2026-115916 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-26 |
| CVE-2026-115914 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-26 |
| CVE-2026-115913 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-26 |
| CVE-2026-115911 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-26 |
| CVE-2026-116181 | 6.5 | MEDIUM | vllm | vLLM from 0.20.1 before 0.29.0 contains an improper input validation vulnerability that allows remote attackers to crash the engine by supplying out-of-vocabulary stop_token_ids. | 2026-10-25 |
| CVE-2026-116178 | 5.9 | MEDIUM | vllm | vLLM from 0.20.0 before 0.30.0 contains a denial of service vulnerability that allows remote attackers to crash the engine by sending inconsistent multimodal metadata to the disaggregated /inference/v1/generate endpoint. | 2026-10-25 |
| CVE-2026-116175 | 6.5 | MEDIUM | vllm | vLLM from 0.19.0 before 0.24.0 contains a decompression bomb vulnerability because the AudioMediaIO multimodal path calls load_audio without the max_duration_s decoded-duration limit. | 2026-10-25 |
| CVE-2026-116174 | 5.9 | MEDIUM | vllm | vLLM from 0.22.0 before 0.24.0 contains an improper input validation vulnerability in the Rust gRPC Generate and GenerateStream service that forwards unchecked prompt token IDs to EngineCore. | 2026-10-25 |
| CVE-2026-116173 | 5.9 | MEDIUM | vllm | vLLM from 0.22.0 before 0.24.0 contains a denial of service vulnerability that allows remote attackers to terminate EngineCore by bypassing logprob_token_ids limits in the Rust gRPC frontend. | 2026-10-25 |
| CVE-2026-116172 | 5.9 | MEDIUM | vllm | vLLM from 0.22.0 before 0.24.0 contains an improper input validation vulnerability in the Rust gRPC converter that allows remote attackers to bypass max_logprobs limits via CandidateTokens.top_n. | 2026-10-25 |
| CVE-2026-114054 | 6.5 | MEDIUM | vllm | vLLM before 0.31.0 contains a cache aliasing vulnerability that allows authenticated attackers to collide shared cache keys by supplying caller-controlled values such as EXIF ImageID or cache_salt. | 2026-10-21 |
| CVE-2026-111991 | 6.3 | MEDIUM | onnx | A security vulnerability has been detected in WongKinYiu yolov9 0.1. | 2026-10-18 |
| CVE-2026-108962 | 6.3 | MEDIUM | comfyui | A vulnerability has been found in storyicon comfyui_segment_anything up to ab6395596399d5048639cdab7e44ec9fae857a93. | 2026-10-13 |
| CVE-2026-109311 | 6.5 | MEDIUM | vllm | vLLM before 0.29.0 passes uploaded audio to the PyAV/FFmpeg av.open() fallback with only a file-size check, exposing FFmpeg's IAMF parser heap overflow. | 2026-10-12 |
| CVE-2026-109296 | 4.3 | MEDIUM | ollama | Open WebUI before 0.11.4 contains a missing authorization vulnerability that allows authenticated users to reach disabled model provider backends by supplying an explicit backend index. | 2026-10-12 |
| CVE-2026-108748 | 5.3 | MEDIUM | langchain | Quarkus LangChain4j 1.9.0 through 1.14.1 contains a missing release of memory vulnerability in the chat-scopes WebSocket /_chat/routes endpoint that allows unauthenticated remote clients to exhaust server memory. | 2026-10-11 |
| CVE-2026-105922 | 4.3 | MEDIUM | vllm | A security flaw has been discovered in vllm-project vLLM up to 0.31.0. | 2026-10-06 |
| CVE-2026-105775 | 4.3 | MEDIUM | vllm | A security vulnerability has been detected in vllm-project vLLM up to 0.31.0. | 2026-10-06 |
| CVE-2026-105760 | 5.3 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-05 |
| CVE-2026-105759 | 5.9 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-05 |
| CVE-2026-105758 | 5.3 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-05 |
| CVE-2026-105757 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-05 |
| CVE-2026-105756 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-05 |
| CVE-2026-105755 | 4.2 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-05 |
| CVE-2026-105754 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-05 |
| CVE-2026-105753 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-05 |
| CVE-2026-103241 | 5.3 | MEDIUM | vllm | A flaw has been found in vllm-project vLLM up to 0.26.0. | 2026-09-30 |
| CVE-2026-101861 | 4.1 | MEDIUM | langchain | Langflow 1.0.16 before 1.12.0 and 0.0.94 before 1.12.0 contain an unsafe eval() vulnerability in schema.py that allows authenticated attackers to achieve code execution by placing a Python object with a malicious __repr__ method into compon | 2026-09-28 |
| CVE-2026-100654 | 6.5 | MEDIUM | vllm | vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model voca | 2026-09-26 |
| CVE-2026-100653 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-09-26 |
| CVE-2026-100652 | 5.9 | MEDIUM | vllm | vLLM versions 0.22.0 through 0.23.0 fail to validate stop_token_ids against vocabulary bounds in Rust HTTP and gRPC frontends, allowing out-of-vocabulary token IDs to reach MinTokensLogitsProcessor. | 2026-09-26 |
| CVE-2026-100651 | 6.5 | MEDIUM | vllm | vLLM before 0.29.0 fails to enforce decoder prompt-length validation on the disaggregated serving endpoint /inference/v1/generate. | 2026-09-26 |
| CVE-2026-100650 | 6.5 | MEDIUM | vllm | vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prom | 2026-09-26 |
| CVE-2026-100648 | 5.3 | MEDIUM | vllm | vllm before 0.29.0 fails to enforce VLLM_MAX_AUDIO_CLIP_FILESIZE_MB limit in multimodal chat audio decoding, allowing unauthenticated clients to bypass file size restrictions. | 2026-09-26 |
| CVE-2026-100647 | 5.3 | MEDIUM | vllm | vLLM versions before 0.29.0 contain a denial-of-service vulnerability in the cache_salt parameter accepted on OpenAI-compatible and Anthropic API endpoints, which lacks maximum length validation and is processed on the single EngineCore sch | 2026-09-26 |
| CVE-2026-97869 | 4.1 | MEDIUM | langchain | A flaw has been found in langchain4j up to 1.5.3-beta10/1.11.10-beta18/1.18.1-beta27. | 2026-09-25 |
| CVE-2026-85709 | 5.3 | MEDIUM | ollama | LightRAG provides simple and fast retrieval-augmented generation. | 2026-09-22 |
| CVE-2026-94625 | 5.3 | MEDIUM | vllm | vLLM through 0.29.0 contains a resource exhaustion vulnerability in MooncakeConnector where rejected prefill requests create ownerless transfer placeholders that are never reclaimed. | 2026-09-21 |
| CVE-2026-94093 | 6.3 | MEDIUM | pytorch, huggingface | A security vulnerability has been detected in DLR-RM stable-baselines3 up to 2.9.0. | 2026-09-20 |
| CVE-2026-69147 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-09-16 |
| CVE-2026-57173 | 6.5 | MEDIUM | vllm | vLLM is an inference and serving engine for large language models. | 2026-09-16 |
| CVE-2026-92365 | 4.3 | MEDIUM | vllm | A vulnerability was found in vllm-project vllm up to 0.29.0. | 2026-09-16 |
| CVE-2026-92220 | 5.3 | MEDIUM | vllm | A vulnerability was found in vllm-project vLLM 0.26.0/0.27.0. | 2026-09-16 |
| CVE-2026-90878 | 4.3 | MEDIUM | vllm | A vulnerability was determined in vllm-project vLLM up to 0.27.1. | 2026-09-15 |
| CVE-2026-55093 | 6.1 | MEDIUM | tensorflow, onnx | Tract is a tiny, no-nonsense, self-contained TensorFlow and ONNX inference toolkit. | 2026-09-14 |
| CVE-2026-55832 | 6.1 | MEDIUM | tensorflow, onnx | Tract is a tiny, no-nonsense, self-contained TensorFlow and ONNX inference toolkit. | 2026-09-14 |
| CVE-2026-90555 | 6.5 | MEDIUM | vllm | vLLM versions before 0.28.0 fail to validate audio sample rate headers in the transcription endpoint, allowing authenticated clients to bypass duration checks. | 2026-09-12 |
| CVE-2026-90554 | 6.2 | MEDIUM | vllm | vLLM versions >=0.10.2 and <0.28.0 do not apply any audio decode-size or duration limit when extracting audio from video input for NanoNemotronVL models. | 2026-09-12 |
| CVE-2026-86289 | 4.3 | MEDIUM | ollama | A vulnerability was found in Ollama up to 0.31.1. | 2026-09-07 |
| CVE-2026-86288 | 6.3 | MEDIUM | triton | A vulnerability has been found in ModelCloud GPTQModel up to 7.2.0. | 2026-09-07 |
| CVE-2026-82637 | 5.3 | MEDIUM | gradio | browser-use web-ui versions 2.0.0 through 3.0.0 fail to validate browser settings paths in run_agent_task, allowing attackers to create directories at arbitrary locations by supplying absolute paths to save_recording_path, save_trace_path, | 2026-08-30 |
| CVE-2026-79785 | 5.9 | MEDIUM | pytorch, onnx | X-AnyLabeling's model downloader disabled TLS certificate verification. | 2026-08-25 |
| CVE-2026-78684 | 5.3 | MEDIUM | vllm | vLLM before 0.27.0 fails to properly classify DeepStream as a GPU backend and omits pixel-limit enforcement in its decode path. | 2026-08-25 |
| CVE-2026-47630 | 5.5 | MEDIUM | triton | NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker could cause an absolute path traversal. | 2026-08-18 |
| CVE-2026-47606 | 6.5 | MEDIUM | triton | NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker could cause an absolute path traversal. | 2026-08-18 |
| CVE-2026-115917 | 3.4 | LOW | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-26 |
| CVE-2026-114049 | 3.7 | LOW | vllm | vLLM before 0.31.0 contains a cross-request mask leak in InputBatch.condense() that allows remote attackers to constrain other clients' outputs by leaving stale allowed_token_ids rows. | 2026-10-21 |
| CVE-2026-105752 | 3.1 | LOW | vllm | vLLM is an inference and serving engine for large language models. | 2026-10-05 |
| CVE-2026-100649 | 3.7 | LOW | vllm | vLLM before 0.29.0 contains a resource-limit bypass vulnerability in PyNvVideoCodec decoder allocation where sampler subclass shadowing allows independent counter increments. | 2026-09-26 |
| CVE-2026-93989 | 3.1 | LOW | vllm | vLLM through 0.29.0 fails to properly validate bad_words token indices against the model's generation output width in SamplingParams.update_from_tokenizer(). | 2026-09-19 |
| CVE-2026-93841 | 3.7 | LOW | vllm, triton | vLLM through 0.29.0 contains a memory corruption vulnerability in the Triton _bincount_kernel where prompt token IDs index the penalty prompt-presence bitset without bounds checking against vocabulary size. | 2026-09-18 |
| CVE-2026-93840 | 3.7 | LOW | vllm | vLLM before 0.29.0 validates allowed_token_ids against tokenizer length instead of model output logits width in SamplingParams._validate_allowed_token_ids(). | 2026-09-18 |
| CVE-2026-90713 | 3.3 | LOW | vllm | A security flaw has been discovered in vllm-project vLLM up to 0.29.0. | 2026-09-14 |
| CVE-2026-63632 | 3.3 | LOW | onnx | Open Neural Network Exchange (ONNX) is an open standard for machine learning interoperability. | 2026-08-18 |
| CVE-2026-123790 | — | UNKNOWN | comfyui | ComfyUI-Easy-Use version 1.3.6 contains a path traversal vulnerability in the `save_preview()` handler for POST `/easyuse/save/{name}`. | 2026-11-12 |
| CVE-2026-83494 | — | UNKNOWN | pytorch | A flaw was found in the training-operator. | 2026-11-06 |
| CVE-2026-103663 | — | UNKNOWN | ollama | Ollama is vulnerable to path traversal in the `/api/pull` endpoint due to insufficient validation of layer digests by the `digestToPath` function. | 2026-10-08 |
| CVE-2026-106119 | — | UNKNOWN | langchain | LangChain is a framework for building LLM-powered applications. | 2026-10-06 |
| CVE-2026-105799 | — | UNKNOWN | langchain | LangChain is a framework for building LLM-powered applications. | 2026-10-06 |
| CVE-2026-62997 | — | UNKNOWN | pytorch | Kedro-Datasets provides data connectors for Kedro. | 2026-09-16 |
Related: Prompt injection · Jailbreak · AI safety · Software coverage · Policy coverage
Data is mirrored from the public NIST NVD 2.0 API and refreshed weekly. Severity and CVSS reflect the highest score recorded across CVSS v3.x metrics. This page is a reference aid, not security advice; always consult the upstream advisory.