Google DeepMind Researchers Propose CaMeL: A Robust Defense that Creates a Protective System Layer around the LLM, Securing It even when Underlying Models may be Susceptible to Attacks