LoLLMS respond_request() Unauthenticated Access

The vulnerability of the respondrequest function in the system for launching and managing large language models of LoLLMS is related to deficiencies in the authentication process. Exploiting this vulnerability could allow a malicious actor, operating remotely, to gain unauthorized access to...

LoLLMS create_post XSS

The vulnerability of the createpost function in the system for launching and managing large language models of LoLLMS Lord of Large Language Multimodal Systems is related to the lack of measures taken to protect the web page structure. Exploiting this vulnerability allows a malicious actor to...

CVE-2026-31229

The Adversarial Robustness Toolbox (ART) versions thru 1.20.1 are vulnerable to an insecure deserialization (CWE-502) flaw within the Kubeflow component 's model loading functionality. The root cause is the use of torch.load() to process model weights (e.g., model.pt) without the security-restric...

CVE-2026-31217

The optimate AI model optimization tool library (versions prior to commit a6d302f912b481c94370811af6b11402f51d377f ) is vulnerable to arbitrary code execution. The _load_model() function in the neural_magic_training.py script fails to validate or sanitize a module.py file when a directory path is...

CVE-2026-31222

The Snorkel library prior to v0.10.0 contains an insecure deserialization vulnerability (CWE-502) in Trainer.load(), where model checkpoints are loaded with torch.load() without weights_only=True. This allows deserialization of arbitrary Python objects via Pickle, enabling remote code execution w...

CTFusion: A CTF-Based Benchmark for LLM Agent Evaluation

Recent advances in Large Language Models LLMs have enabled agentic systems for complex, multi-step tasks; cybersecurity is emerging as a prominent application. To evaluate such agents, researchers widely adopt Capture The Flag CTF benchmarks. However, current CTF benchmarks reuse existing...

LLM security vulnerabilities

LLM is a multi-model large language model command-line interaction tool developed by Simon Willison. Versions of LLM 0.27.1 and earlier contain security vulnerabilities. These vulnerabilities stem from the use of the --functions command-line parameter to directly execute unsafe code using the exe...

SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces

Reusable skills are becoming a common interface for extending large language model agents, packaging procedural guidance with access to files, tools, memory, and execution environments. However, this modularity introduces attack surfaces that are largely missed by existing safety evaluations: eve...

OptiMate security vulnerabilities

OptiMate is an AI model optimization tool library developed by Nebuly. There is a security vulnerability in OptiMate. This vulnerability stems from the loadmodel function in the neuralmagictraining.py script, which loads model files using torch.load, without enabling the weightsonly=True paramete...

Mamba security vulnerabilities

Mamba is a state-space model for linear time series modeling, open-sourced by State-Spaces. Versions of Mamba 2.2.6 and earlier contained security vulnerabilities. These vulnerabilities stemmed from the MambaLMHeadModel.frompretrained method, which used torch.load to load weight files without...

CosyVoice security vulnerabilities

CosyVoice is an open-source voice generation and AI voice cloning platform developed by FunAudioLLM. CosyVoice has a security vulnerability. This vulnerability arises from the model loading process, where the .pt files in the user-specified directory are loaded using torch.load, without enabling...

Adversarial Robustness Toolbox Security Vulnerabilities

Adversarial Robustness Toolbox is an open-source machine learning security defense and evaluation tool developed by Trusted-AI. Versions of Adversarial Robustness Toolbox 1.20.1 and earlier contained security vulnerabilities. These vulnerabilities stemmed from the model loading function in the...

VMware Spring AI Security Vulnerabilities

VMware Spring AI is a development framework created by VMware Corporation in the Spring ecosystem, which integrates artificial intelligence and large language model capabilities. VMware Spring AI has a security vulnerability. This vulnerability allows malicious users to manipulate the behavior of...

CVE-2026-31238

The Ludwig framework thru 0.10.4 is vulnerable to insecure deserialization CWE-502 in its model serving component. When starting a model server with the ludwig serve command, the framework loads model weight files using torch.load without enabling the security-restrictive weightsonly=True...

CVE-2026-31219

The optimate project, an AI model optimization tool library by Nebuly, contains an insecure deserialization (CWE-502) vulnerability in the _load_model() function of the neural_magic_training.py script (commit a6d302f912b481c94370811af6b11402f51d377f). The issue occurs because the function uses to...

CVE-2026-31238

The Ludwig framework thru 0.10.4 is vulnerable to insecure deserialization CWE-502 in its model serving component. When starting a model server with the ludwig serve command, the framework loads model weight files using torch.load without enabling the security-restrictive weightsonly=True...

CVE-2026-31217

The loadmodel function in the neuralmagictraining.py script of the optimate project in commit a6d302f912b481c94370811af6b11402f51d377f 2024-07-21 allows arbitrary code execution. When a user supplies a directory path via the --model command-line argument, the function reads a module.py file from...

CVE-2026-31217: Improper Control of Generation of Code

The loadmodel function in the neuralmagictraining.py script of the optimate project in commit a6d302f912b481c94370811af6b11402f51d377f 2024-07-21 allows arbitrary code execution. When a user supplies a directory path via the --model command-line argument, the function reads a module.py file from...

CVE-2026-41713: Improper Neutralization of Special Elements Used in a Template Engine

A malicious user could craft input that is stored in conversation memory and later interpreted by the model in an unintended way. Applications using the affected advisor with user-controlled input may be susceptible to manipulation of model behavior across conversation turns...

PT-2026-40007

A malicious user could craft input that is stored in conversation memory and later interpreted by the model in an unintended way. Applications using the affected advisor with user-controlled input may be susceptible to manipulation of model behavior across conversation turns...