AI Sec Watch

MLflow has a vulnerability (CVE-2024-27134) where directories have overly permissive access settings, allowing a local attacker to gain elevated permissions through a ToCToU attack (a race condition where an attacker exploits the gap between when a program checks permissions and when it uses a resource). This only affects code using the spark_udf() MLflow API.

A security flaw in Hugging Face Transformers allows attackers to run arbitrary code (RCE, remote code execution) on a user's computer by tricking them into opening a malicious file or visiting a malicious webpage. The vulnerability happens because the software doesn't properly validate data when loading model files, allowing untrusted data to be deserialized (converted from storage format back into a running program).

A vulnerability in Hugging Face Transformers' MaskFormer model allows attackers to run arbitrary code (RCE, or remote code execution) on a user's computer if they visit a malicious webpage or open a malicious file. The flaw occurs because the model file parser doesn't properly validate user-supplied data before deserializing it (converting saved data back into working code), allowing attackers to inject and execute malicious code.

Hugging Face Transformers MobileViTV2 has a vulnerability where attackers can execute arbitrary code (running commands they choose) by tricking users into visiting malicious pages or opening malicious files that contain specially crafted configuration files. The flaw happens because the software doesn't properly check (validate) data before deserializing it (converting it from stored format back into usable code), allowing untrusted data to be executed.

LLama Factory, a tool for fine-tuning large language models (AI systems trained on specific tasks or data), has a critical vulnerability that lets attackers run arbitrary commands on the computer running it. The flaw comes from unsafe handling of user input, specifically using a Python function called `Popen` with `shell=True` (a setting that interprets input as system commands) without checking or cleaning the input first.

MarkUs (a web application for student assignment submission and grading) has a vulnerability in versions before 2.4.8 that allows authenticated instructors to write files anywhere on the web server, potentially leading to remote code execution (the ability to run commands on a system from a distance). This happens because the file upload methods don't properly restrict where files can be saved.

This is the official 2025 release of the OWASP Top 10 for Large Language Model Applications, which is a ranked list of the most critical security risks affecting AI systems. The document provides guidance on the biggest threats that developers should be aware of when building or using LLM-based applications (software built around large language models, which are AI systems trained on vast amounts of text).

A WordPress plugin called Sage AI (which provides chatbots, GPT-4 article generation, and image creation features) has a vulnerability (CVE-2024-52384) that allows unrestricted uploading of dangerous file types, enabling attackers to upload web shells (malicious scripts that give attackers control of a web server). This vulnerability affects all versions up to and including version 2.4.9.

CVE-2024-52383 is a missing authorization vulnerability (a flaw where the software fails to check if a user has permission to perform an action) in the KCT Ai Auto Tool Content Writing Assistant plugin for WordPress, affecting versions up to 2.1.2. This vulnerability allows attackers to exploit incorrectly configured access control (permission settings) to gain unauthorized access.