llama.cpp 安全漏洞

llama.cpp is a multimodal model by the individual developer Georgi Gerganov. A security vulnerability exists in versions of llama.cpp prior to b5721, which stems from the presence of signed and unsigned integer overflows in the tokenizer implementation, which could lead to a heap overflow...

Watermarking Autoregressive Image Generation

Watermarking the outputs of generative models has emerged as a promising approach for tracking their provenance. Despite significant interest in autoregressive image generation models and their potential for misuse, no prior work has attempted to watermark their outputs at the token level. In thi...

Path Traversal in Tokenizer Conversion Script

The script for converting slow tokenizers is vulnerable to a Path Traversal attack via the --checkpointname command-line argument. This allows an attacker to create files outside of the intended dumppath directory. Vulnerable Code Location: The vulnerability is located in the logic for converting...

Rubber Mallet: a Study of High Frequency Localized Bit Flips and Their Impact on Security

The increasing density of modern DRAM has heightened its vulnerability to Rowhammer attacks, which induce bit flips by repeatedly accessing specific memory rows. This paper presents an analysis of bit flip patterns generated by advanced Rowhammer techniques that bypass existing hardware defenses...

Regular expression Denial of Service - ReDoS

Description A regular expression denial of service ReDoS vulnerability has been identified in the Hugging Face Transformers library's MarianTokenizer. The vulnerability exists in the removelanguagecode method of the MarianTokenizer class, which processes text to remove language codes. The method...

Amazon Linux 2023 : ecs-init (ALAS2023-2025-1011)

It is, therefore, affected by a vulnerability as referenced in the ALAS2023-2025-1011 advisory. The tokenizer incorrectly interprets tags with unquoted attribute values that end with a solidus character / as self-closing. When directly using Tokenizer, this can result in such tags incorrectly bei...

Medium: ecs-init

Issue Overview: The tokenizer incorrectly interprets tags with unquoted attribute values that end with a solidus character / as self-closing. When directly using Tokenizer, this can result in such tags incorrectly being marked as self-closing, and when using the Parse functions, this can result i...

Medium: ecs-init

Issue Overview: The tokenizer incorrectly interprets tags with unquoted attribute values that end with a solidus character / as self-closing. When directly using Tokenizer, this can result in such tags incorrectly being marked as self-closing, and when using the Parse functions, this can result i...

Medium: ecs-init

Issue Overview: The tokenizer incorrectly interprets tags with unquoted attribute values that end with a solidus character / as self-closing. When directly using Tokenizer, this can result in such tags incorrectly being marked as self-closing, and when using the Parse functions, this can result i...

GradEscape: a Gradient-Based Evader against AI-Generated Text Detectors

In this paper, we introduce GradEscape, the first gradient-based evader designed to attack AI-generated text AIGT detectors. GradEscape overcomes the undifferentiable computation problem, caused by the discrete nature of text, by introducing a novel approach to construct weighted embeddings for t...

Important: soci-snapshotter

Issue Overview: The net/http package accepted data in the chunked transfer encoding containing an invalid chunk-size line terminated by a bare LF. When used in conjunction with a server or proxy which incorrectly interprets a bare LF in a chunk extension as part of the extension, this could permi...

Important: nerdctl

Issue Overview: The net/http package accepted data in the chunked transfer encoding containing an invalid chunk-size line terminated by a bare LF. When used in conjunction with a server or proxy which incorrectly interprets a bare LF in a chunk extension as part of the extension, this could permi...

Amazon Linux 2 : nerdctl (ALAS-2025-2863)

The version of nerdctl installed on the remote host is prior to 2.0.5-1. It is, therefore, affected by multiple vulnerabilities as referenced in the ALAS2-2025-2863 advisory. The net/http package accepted data in the chunked transfer encoding containing an invalid chunk-size line terminated by a...

CBL Mariner 2.0 Security Update: cf-cli / cni-plugins / containerized-data-importer / docker-compose / kubevirt / sriov-network-device-plugin (CVE-2025-22872)

The version of cf-cli / cni-plugins / containerized-data-importer / docker-compose / kubevirt / sriov-network-device- plugin installed on the remote CBL Mariner 2.0 host is prior to tested version. It is, therefore, affected by a vulnerability as referenced in the CVE-2025-22872 advisory. - The...

CVE-2025-46560

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.8.0 and prior to 0.8.5 are affected by a critical performance vulnerability in the input preprocessing logic of the multimodal tokenizer. The code dynamically replaces placeholder tokens...

CVE-2025-46560 vLLM phi4mm: Quadratic Time Complexity in Input Token Processing​ leads to denial of service

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.8.0 and prior to 0.8.5 are affected by a critical performance vulnerability in the input preprocessing logic of the multimodal tokenizer. The code dynamically replaces placeholder tokens...

CVE-2025-46560

CVE-2025-46560 affects vLLM 0.8.0–0.8.4, where the multimodal tokenizer’s input preprocessing uses placeholder tokens replaced by repeated tokens. The replacement logic relies on inefficient list concatenation, yielding quadratic time complexity (O(n²)) and enabling resource exhaustion via crafte...

CVE-2025-46560 vLLM phi4mm: Quadratic Time Complexity in Input Token Processing​ leads to denial of service

vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.8.0 and prior to 0.8.5 are affected by a critical performance vulnerability in the input preprocessing logic of the multimodal tokenizer. The code dynamically replaces placeholder tokens...

GHSA-VC6M-HM49-G9QG vLLM: Quadratic Time Complexity in Input Token Processing​ leads to denial of service

Summary A critical performance vulnerability has been identified in the input preprocessing logic of the multimodal tokenizer. The code dynamically replaces placeholder tokens e.g., , with repeated tokens based on precomputed lengths. Due to ​​inefficient list concatenation operations​​, the...

Regular Expression Denial of Service (ReDoS)

Overview transformers is a State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow Affected versions of this package are vulnerable to Regular Expression Denial of Service ReDoS in the tokenizationgptneoxjapanese.py file, within the SubWordJapaneseTokenizer class. The regex is designed ...