AI Sec Watch

The mamba language model framework (versions up to 2.2.6) has a vulnerability in how it loads pre-trained models from HuggingFace Hub (a platform where AI models are shared). When loading models, it uses an unsafe method called torch.load() without the weights_only=True security parameter, which allows attackers to sneak malicious code into model files. An attacker could upload a compromised model to HuggingFace Hub, and when someone downloads and loads it, the attacker's code runs on their computer.

The Ludwig framework (a machine learning tool) versions up to 0.10.4 has a vulnerability where it unsafely loads model files using a method that can execute arbitrary code. When someone runs the ludwig serve command to host a model, an attacker can provide a malicious model file that tricks the system into running their code, potentially taking over the server.

The CosyVoice project has an insecure deserialization vulnerability (CWE-502, a weakness where untrusted data is converted back into executable objects) in how it loads model files. When users load model files (.pt files, which are PyTorch model formats) from a directory they specify, the code uses torch.load() without security protections, allowing attackers to execute arbitrary code by hiding malicious instructions in crafted model files that get executed when loaded.

The Adversarial Robustness Toolbox (ART) up to version 1.20.1 has a vulnerability in its Kubeflow component where it uses eval() (a function that runs text as if it were code) unsafely to process command-line arguments like --clip_values and --input_shape. An attacker can inject malicious Python code through these arguments, which will execute when eval() processes them, potentially giving the attacker full control over the system running ART if they can control those arguments.

The Adversarial Robustness Toolbox (ART) version 1.20.1 and earlier has a vulnerability in how it loads AI model files, specifically in its Kubeflow component (a system for running machine learning workflows). When loading model weights using torch.load() without the weights_only=True security parameter, the software deserializes arbitrary Python objects via Pickle (a Python serialization library), allowing attackers to execute malicious code by uploading a crafted model file or manipulating the model location parameter.

This article describes a legal dispute where OpenAI CEO Sam Altman testified that Elon Musk tried to gain control of OpenAI, the company behind ChatGPT, including suggesting the company could pass to his children when he dies. Altman and other co-founders rejected Musk's control demands because they believed that no single person should control AGI (artificial general intelligence, an AI system that could outperform humans at most tasks). Musk eventually left OpenAI in 2018 and declined to invest when the company restructured into a for-profit entity.

LLM version 0.32a2 is a command-line tool update that adds support for OpenAI's newer /v1/responses endpoint, which allows reasoning-capable models to show their thinking process across multiple steps. Users can now see summarized reasoning tokens displayed in a different color, or hide them using the -R or --hide-reasoning flags.

OpenAI CEO Sam Altman testified that Elon Musk's management style, which involved ranking researchers and cutting staff aggressively, caused significant damage to the company's culture and morale. Altman stated that Musk's approach was incompatible with running a successful research lab, highlighting a clash between different management philosophies at the AI startup.

Exaforce, a company building an agentic SOC (security operations center, where AI agents automate security tasks) platform, raised $125 million in funding to expand its technology. The platform uses autonomous AI agents called Exabots and a real-time knowledge graph (a connected database of security events and relationships) to automatically detect, investigate, and respond to security threats in cloud and SaaS environments without needing traditional SIEM (security information and event management, a tool that collects and analyzes security data) rules.