AI Software
AI software refers to the specialized tools and building blocks used to create, run, and manage artificial intelligence. Just as a chef needs specific ingredients and kitchen tools to cook, AI developers use these software packages to build their systems. Because these tools are so popular, attackers sometimes try to hide malicious code inside them or create fake versions to trick developers into downloading them.
AI software encompasses the libraries, frameworks, and DevOps tooling essential for the development, training, and deployment of AI models. This includes deep learning frameworks like PyTorch, generative AI integration tools, and infrastructure for managing AI workflows. Security risks arise when adversaries target these packages—or their underlying dependency chains—through techniques like dependency confusion, typosquatting, or compromising legitimate repositories to inject malicious code into the AI development lifecycle.
AI software constitutes the ecosystem of software packages, deep learning frameworks (e.g., PyTorch, TensorFlow, JAX), generative AI integration frameworks (e.g., LangChain), and AI DevOps orchestration tools that facilitate the AI lifecycle. Adversaries target this software layer by exploiting the supply chain, including the compromise of upstream dependencies, the distribution of malicious packages via namesquatting or slopsquatting (targeting LLM-hallucinated entities), and the weaponization of configuration files or distributed artifacts like Jupyter notebooks. These threats manifest through techniques such as package compromise, malicious version updates (rug pulls), and the exploitation of AI-specific integration points to achieve unauthorized code execution within the development or production environment.