AI Sec Watch

Retrieval-based-Voice-Conversion-WebUI, a voice changing tool, has a vulnerability in versions 2.2.231006 and earlier where unsafe deserialization (loading data in a way that can execute malicious code) allows attackers to run code remotely. The problem occurs because the software takes user input for model file paths and loads them using torch.load without proper safety checks, enabling RCE (remote code execution, where attackers can run commands on the affected system).

Retrieval-based-Voice-Conversion-WebUI, a voice-changing tool, has a vulnerability in versions 2.2.231006 and earlier where user input for model file paths is passed unsafely to torch.load (a function that loads saved AI models). This unsafe deserialization (loading data from untrusted sources without checking it first) can allow attackers to run arbitrary code on the system.

Retrieval-based-Voice-Conversion-WebUI, a voice-changing framework, has a critical vulnerability in versions 2.2.231006 and earlier where unsafe deserialization (loading data from untrusted sources without checking it first) can occur. An attacker can exploit this by providing a malicious file path that gets loaded using torch.load, which can lead to RCE (remote code execution, where an attacker runs commands on a system they don't own).

Retrieval-based-Voice-Conversion-WebUI, a voice changing tool based on VITS (a voice synthesis model), has a vulnerability in versions 2.2.231006 and earlier where user-supplied file paths are loaded directly using torch.load (a function that can execute code when loading files), allowing attackers to run arbitrary code on the system. This happens because the ckpt_path1 variable accepts untrusted input and passes it unsafely to a model-loading function.

ChatGPT has two memory features: saved memories (which users can manage) and chat history (a newer feature that builds a profile over time without user visibility or control). The chat history feature doesn't search past conversations but maintains recent chat history and learns user preferences, though the implementation details are not publicly documented, and users cannot inspect or modify what the system learns about them unless they use prompt hacking (manipulating the AI's instructions to reveal hidden information).

The Model Context Protocol (MCP) is a system that lets AI applications discover and use external tools from servers at runtime (while the program is running). However, MCP has a security weakness: because servers can send instructions through the tool descriptions, they can perform prompt injection (tricking an AI by hiding instructions in its input) to control the AI client, making servers more powerful than they should be.

AI regulatory sandboxes are controlled testing environments where companies can develop and test AI systems with guidance from regulators before releasing them to the public, as required by the EU AI Act (EU's new rules for artificial intelligence). These sandboxes help companies understand what regulations they must follow, protect them from fines if they follow official guidance, and make it easier for small startups to enter the market. Each EU Member State must create at least one sandbox by August 2, 2026, though different countries are taking different approaches to organizing them.

CVE-2025-46567 is a critical vulnerability in LLaMA-Factory (a tool for fine-tuning large language models) that exists before version 1.0.0. The vulnerability is in the `llamafy_baichuan2.py` script, which unsafely loads user-supplied files using `torch.load()` (a function that deserializes, or reconstructs, Python objects from saved data), allowing attackers to execute arbitrary commands by crafting a malicious file.

vLLM (a system for running large language models efficiently) versions 0.8.0 through 0.8.4 have a critical performance bug in how it processes multimodal input (text, images, audio). The bug uses an inefficient algorithm (quadratic time complexity, meaning it slows down exponentially as input size grows) when replacing placeholder tokens (special markers like <|audio_|> that get expanded into repeated tokens), which allows attackers to crash or freeze the system by sending specially crafted malicious inputs.