Researchers at Bloomberg have published two papers highlighting risks associated with the deployment of Generative AI (GenAI) systems, particularly in sensitive sectors such as capital markets and financial services. One study, titled "RAG LLMs Are NOT Safer," reveals that Retrieval-Augmented Generation (RAG) frameworks, which integrate retrieval of relevant external context with generative language models, can paradoxically reduce the safety of large language models (LLMs). The findings indicate that even when both the AI models and the documents they retrieve are considered "safe," the combined system can still produce unsafe outputs. Traditional red-teaming methods, commonly used to test AI safety, were found to be less effective in the context of RAG. Another paper, "Understanding & Mitigating Risks of Generative AI in Financial Services," critiques existing guardrail solutions for their inadequacy in detecting domain-specific risks and proposes the first finance-specific AI content risk taxonomy to better address these challenges. These insights underscore the complexity of safely architecting and testing AI systems that utilize LLMs and RAG techniques, emphasizing the need for domain-aware safety measures. Additional research in the AI community is exploring secure, multifaceted RAG frameworks to mitigate issues such as limited retrieval scope, data security, and operational costs in enterprise applications.