Prompt injection attacks manipulate LLMs by introducing malicious commands into free text inputs, posing a significant threat to cybersecurity and potentially leading to unauthorized activities or data leaks.