Imagine typing a simple phrase like "walk forward" and having a computer generate a realistic human animation doing just that. This is the power of text-to-motion (T2M) AI. But what if someone could manipulate these systems to generate harmful or inappropriate motions? Researchers have developed ALERT-Motion, a system that uses large language models (LLMs) to create "adversarial attacks" on T2M models. These attacks subtly alter text prompts to trick the AI into generating targeted motions—imagine changing "wave hello" into a motion that looks more like a threatening gesture. Unlike previous methods that used clunky, easily detectable changes to text, ALERT-Motion leverages the LLM's understanding of language and motion to craft subtle, yet effective, adversarial prompts. This is done through two main components: an adaptive dispatching module that refines and searches for adversarial prompts and a module that uses motion data to steer the LLM toward generating the desired motion. This research highlights a growing concern in the field of AI-generated content: the potential for misuse. While T2M technology has exciting applications in areas like animation and virtual reality, safeguarding these systems against manipulation is crucial. Future research will focus on developing defenses against these attacks, perhaps by training T2M models on more diverse datasets or by integrating techniques from adversarial training in natural language processing. As AI systems become more sophisticated, so too must our understanding of their vulnerabilities and our efforts to protect against exploitation.