Zero AI outputs from models like GPT-5.4 and Claude Opus 4.6 passed a quality review by 500 investment bankers. The professionals flagged the results as imprecise or incorrect. Despite the failure, over half will use these tools for initial drafts. This highlights a persistent gap between LLM capabilities and high-stakes financial precision.