Prompt injection attacks, like the recent Policy Puppetry Attack by HiddenLayer, pose significant risks to large language models (LLMs) by exploiting vulnerabilities to make them produce harmful outputs or reveal sensitive information. This attack bypasses safety measures across various models by using a cleverly constructed prompt that combines role-playing, pseudo-code, and encoded language to mislead the AI into executing unintended commands. As these models are increasingly integrated into critical sectors such as healthcare and finance, the need for robust defenses against such attacks becomes imperative. Traditional alignment methods, such as Reinforcement Learning from Human Feedback, are insufficient against novel adversarial strategies, highlighting the importance of continuous monitoring and detection systems like Deepchecks. Deepchecks offers proactive detection by evaluating prompt safety, helping identify and respond to malicious inputs effectively. This approach allows for improved red-teaming efforts and system adjustments to enhance the resilience of AI models against evolving threats.