Generative AI and Large Language Models (LLMs) offer exciting possibilities for automating tasks in spreadsheets, from simple formula creation to complex financial modeling. But can we truly trust these powerful tools with our data and decisions? New research explores the critical concept of trust in AI-powered spreadsheets, highlighting potential pitfalls and proposing a framework for evaluating their trustworthiness. Imagine asking an AI to build a complex spreadsheet formula. It delivers instantly, but how do you know it’s correct? The problem is that LLMs are prone to “hallucinations”—generating outputs that are grammatically correct but factually wrong. This, coupled with potential biases embedded in their training data and the complexity of user prompts, creates a significant challenge for trusting AI-generated spreadsheet content. To address this, researchers propose a “Transparency and Trustworthiness Framework” based on two core dimensions: transparency and dependability. Transparency emphasizes explainability (understanding the AI's reasoning) and visibility (inspecting its underlying algorithms). Dependability focuses on reliability (consistent accuracy) and ethical considerations (bias and fairness). This framework offers a structured approach to scrutinizing AI-generated formulas, moving beyond blind faith to informed assessment. The research also dives into the sources of errors that erode trust in AI, highlighting how hallucinations are often triggered by uncertainty, negation, or complex reasoning in prompts. User biases, like “magical thinking,” where we overestimate AI capabilities, and “reification,” where we treat abstract models as unquestionable truths, also contribute to the problem. The paper underscores how prompt engineering—the art of crafting effective instructions for AI—plays a vital role in mitigating these issues. The consequences of misplaced trust in automated systems are starkly illustrated through real-world examples, including the Reinhart-Rogoff economic study, the UK’s COVID-19 test-and-trace system, and the Post Office Horizon scandal. These cases underscore how seemingly minor errors can have far-reaching consequences, from influencing national policy to causing significant financial and personal harm. Looking ahead, this research lays the groundwork for building truly trustworthy AI-powered spreadsheets. Further research into planning methods for prompt engineering and agile development techniques like test-driven development promises to improve the reliability of AI-generated formulas. Ultimately, the goal is to empower users to confidently harness the power of AI while maintaining critical oversight and ensuring responsible implementation.