Guidelines to Preventing Artificial Intelligence Hallucinations in Microsoft Fabric
Year: 2026
Type: other
Abstract: This paper examines why Microsoft Copilot in Fabric can generate confident but inaccurate outputs—fabricated values, misread schemas, or unsupported conclusions—and proposes a six-layer, defense-in-depth framework for reducing that risk. Drawing on a narrative synthesis of twenty-four academic and industry sources, the framework combines governance controls, data quality standards, retrieval-augmented grounding, structured prompting, human review, and automated detection across the AI lifecycle. Rather than relying on a single fix, it offers a phased implementation roadmap—starting with data cleanup and prompt discipline, then layering in governance, grounding, and automated safeguards—along with a proposed stress-test protocol for future validation. The result is a research-grounded starting point for deploying Copilot in Fabric with greater accuracy, oversight, and compliance.
Institution: Lumen (United Kingdom)
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7155195483