Digital STEALTH Metric

Digital STEALTH metrics assess the ability of malicious attacks to subtly manipulate machine learning models, particularly large language models (LLMs) and deep learning systems for image and video processing, without detection. Current research focuses on developing and evaluating these attacks, employing techniques like backdoor injections, adversarial examples (including those generated by diffusion models), and gradient manipulation within various architectures (e.g., LSTMs, GPT-2, and convolutional networks). Understanding and quantifying STEALTH is crucial for improving the robustness and security of AI systems across diverse applications, ranging from autonomous vehicles to healthcare and finance.