AI Sec Watch

TensorFlow, an open source platform for machine learning, had an out of bounds read vulnerability (a bug where code tries to access memory it shouldn't) in a component called GRUBlockCellGrad before versions 2.12.0 and 2.11.1. This vulnerability could potentially allow attackers to read sensitive data or crash the system.

CVE-2023-1177 is a path traversal vulnerability (a flaw where an attacker can access files outside the intended directory by using special characters like '..') in MLflow versions before 2.2.1. This weakness allows attackers to potentially read or access files they shouldn't be able to reach on the system.

CVE-2023-1176 is an absolute path traversal vulnerability (a bug where an attacker can access files anywhere on a system by using file paths that start from the root directory) found in MLflow, an open-source platform for managing machine learning experiments, affecting versions before 2.2.2. The vulnerability was discovered and reported through the huntr.dev bug bounty program.

Streamlit, software that converts data scripts into web applications, had a cross-site scripting vulnerability (XSS, where an attacker injects malicious code that runs in a user's browser) in versions 0.63.0 through 0.80.0. An attacker could craft a malicious URL containing JavaScript code, trick a user into clicking it, and the Streamlit server would execute that code in the victim's browser.

Yolo is a tool that uses ChatGPT API (OpenAI's language model accessed through code) to translate natural language questions into shell commands (the text-based interface for controlling a computer) that can be executed automatically. The tool helps users who forget command syntax by converting plain English requests into proper bash, zsh, or PowerShell commands, with a safety feature that shows the command before running it unless the user enables automatic execution.

LiteDB, a lightweight database library for .NET, has a vulnerability in versions before 5.0.13 where it can deserialize (convert data from a format like JSON back into usable objects) untrusted data. If an attacker sends specially crafted JSON to an application using LiteDB, the library may load unsafe objects by using a special `_type` field that tells it what class to create, potentially allowing malicious code execution.

Gradio is a Python library for building AI demo applications, and versions before 3.13.1 accidentally exposed private SSH keys (security credentials that grant system access) when users enabled share links to let others access their apps. This meant anyone connecting to a shared Gradio app could steal the SSH key and access other users' Gradio demos or exploit them further depending on what data or capabilities the app had access to.

CVE-2022-26076 is a vulnerability in Intel's oneAPI Deep Neural Network library (oneDNN, a software framework for machine learning tasks) before version 2022.1 that involves an uncontrolled search path element (a weakness where a program looks for files in directories it shouldn't trust, potentially allowing attackers to substitute malicious files). An authenticated user (someone with login access) could exploit this through local access to gain higher system privileges.

CVE-2023-23382 is a vulnerability in Azure Machine Learning Compute Instance that allows unauthorized access to sensitive information. The vulnerability is related to storing passwords in a recoverable format (CWE-257, meaning passwords are saved in a way that can be converted back to their original form), making it easier for attackers to steal credentials.