How do you handle prompt injection attacks or ensure that an LLM does not hallucinate when summarizing post-trade accounting data at BlackRock?
