Google-Tensor-Chips

Google Tensor Chips — DarWINN NPU Access Direct-access tooling for the Google Tensor NPU DarWINN, reverse-engineered across two generations: Tensor G2 Pixel 7, codename Janeiro and Tensor G3 Pixel 8, codename Rio. Why this exists I was trying to get LLM inference running on the phone and wanted t...

CVE-2026-34159-Vulnerability-Research-Analysis-Detection

CVE-2026-34159-Vulnerability-Research-Analysis-Detection Critical buffer validation bypass in deserializetensor llama.cpp b8492. Null tensor buffer skips bounds check, enabling unauthenticated arbitrary R/W via GRAPHCOMPUTE. ASLR bypass + RCE over TCP 50052. Contains: Technical analysis & exploit...

CVE-2026-7020

CVE-2026-7020 Ollama arbitrary file read via tensor digest path traversal. Exfiltrates SSH host keys and any other file from a reachable Ollama host in three unauthenticated API calls. Writeup: Co-Researcher @gouldnicholas Usage python3 poc.py Tries all three SSH host key types with known sizes a...

EUVD-2026-110041

ONNX: Out-of-bounds read in ConvTranspose shape inference via undersized weight rank...

GHSA-Q97C-P368-75C4 vLLM: Out-of-vocabulary prompt token id passes input validation and triggers a CUDA device-side assert (engine crash / DoS)

For a model whose vocabulary size is an exact multiple of the embedding padding size 64 — e.g. any Llama-2-family model, vocab 32000 — a single request carrying a pre-tokenized prompt with token id equal to the vocab size 32000 passes vLLM's input-id validation but is one past the last embedding...

vLLM: Out-of-vocabulary prompt token id passes input validation and triggers a CUDA device-side assert (engine crash / DoS)

For a model whose vocabulary size is an exact multiple of the embedding padding size 64 — e.g. any Llama-2-family model, vocab 32000 — a single request carrying a pre-tokenized prompt with token id equal to the vocab size 32000 passes vLLM's input-id validation but is one past the last embedding...

CVE-2026-115922

A flaw was found in ONNX. This vulnerability allows an attacker to cause a Denial of Service DoS through excessive memory consumption by providing a specially crafted machine learning model. The issue occurs because tensor materialization routines resize memory arrays based on dimensions declared...

CVE-2024-58399

llama.cpp before b2731 contains an out-of-bounds read vulnerability in llamamodelloader that allows attackers to read memory outside the mapped model file via crafted GGUF tensor offsets. Attackers can supply a crafted or truncated GGUF model file with arbitrary 64-bit tensor offsets to crash the...

UBUNTU-CVE-2024-58399

llama.cpp before b2731 contains an out-of-bounds read vulnerability in llamamodelloader that allows attackers to read memory outside the mapped model file via crafted GGUF tensor offsets. Attackers can supply a crafted or truncated GGUF model file with arbitrary 64-bit tensor offsets to crash the...

CVE-2024-58399

The llama.cpp library (versions prior to b2731 ) contains an out-of-bounds read vulnerability within the llama_model_loader component. An attacker can exploit this by providing a crafted or truncated GGUF model file containing arbitrary 64-bit tensor offsets. This allows the attacker to read memo...

CVE-2024-58399 llama.cpp before b2731 Out-of-Bounds Read via GGUF Tensor Offsets

llama.cpp before b2731 contains an out-of-bounds read vulnerability in llamamodelloader that allows attackers to read memory outside the mapped model file via crafted GGUF tensor offsets. Attackers can supply a crafted or truncated GGUF model file with arbitrary 64-bit tensor offsets to crash the...

EUVD-2024-55826

llama.cpp before b2731 contains an out-of-bounds read vulnerability in llamamodelloader that allows attackers to read memory outside the mapped model file via crafted GGUF tensor offsets. Attackers can supply a crafted or truncated GGUF model file with arbitrary 64-bit tensor offsets to crash the...

CVE-2024-58399 llama.cpp before b2731 Out-of-Bounds Read via GGUF Tensor Offsets

llama.cpp before b2731 contains an out-of-bounds read vulnerability in llamamodelloader that allows attackers to read memory outside the mapped model file via crafted GGUF tensor offsets. Attackers can supply a crafted or truncated GGUF model file with arbitrary 64-bit tensor offsets to crash the...

CVE-2024-58399

llama.cpp before b2731 contains an out-of-bounds read vulnerability in llamamodelloader that allows attackers to read memory outside the mapped model file via crafted GGUF tensor offsets. Attackers can supply a crafted or truncated GGUF model file with arbitrary 64-bit tensor offsets to crash the...

PT-2026-125540

llama.cpp before b2731 contains an out-of-bounds read vulnerability in llama model loader that allows attackers to read memory outside the mapped model file via crafted GGUF tensor offsets. Attackers can supply a crafted or truncated GGUF model file with arbitrary 64-bit tensor offsets to crash t...

Reachable Assertion

Overview vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs Affected versions of this package are vulnerable to Reachable Assertion via parseandvalidateimageinput in deepseekocr.py and deepseekocr2.py, where a zero-sum pixel tensor causes the function to retur...

DEBIAN-CVE-2026-115920

Open Neural Network Exchange ONNX is an open standard for machine learning interoperability. From 1.2.1 until 1.22.0, ConvTranspose shape inference processes a weight tensor with fewer than three dimensions without a kernelshape attribute by deriving an empty effectivekernelshape and indexing it...

UBUNTU-CVE-2026-115920

Open Neural Network Exchange ONNX is an open standard for machine learning interoperability. From 1.2.1 until 1.22.0, ConvTranspose shape inference processes a weight tensor with fewer than three dimensions without a kernelshape attribute by deriving an empty effectivekernelshape and indexing it...

PT-2026-117223

A vulnerability has been found in hankcs HanLP up to 2.1.0-beta.64. The affected element is the function load word2vec as vocab tensor of the file hanlp/utils/torch util.py. The manipulation of the argument vectors.vocab leads to deserialization. The attack is possible to be carried out remotely...

ai.djl/api: Integer overflow in tensor buffer validation in Deep Java Library

A flaw was found in Deep Java Library. Due to an integer overflow during tensor buffer validation, a remote, unauthenticated attacker can send a crafted tensor payload to trigger out-of-bounds memory access. Successful exploitation can allow an attacker to obtain sensitive data from adjacent...