Research · TechCrunch ·

KPMG pulls AI report over hallucinated data

A major consulting firm withdrew a high-profile AI study after discovering that some of its data and conclusions were generated by the very AI systems it was analyzing.

Based on reporting by TechCrunch — analysis by dalili

KPMG has retracted a widely-cited report on enterprise AI adoption after discovering that portions of the data analysis were conducted using AI models, and some of the conclusions were generated by the models rather than human analysis. The report, released in early June, had become a reference point for C-suite discussions about AI investment strategy.

The firm acknowledged that internal QA processes failed to catch that AI-generated content had been mixed into the final deliverable. KPMG's researchers had used Claude and ChatGPT to assist with data processing, but failed to manually verify the outputs before publication.

The incident raises questions about the reliability of AI-assisted analysis in high-stakes consulting work. If a firm specializing in technology cannot reliably distinguish between human-authored and AI-generated content in its own reports, how can enterprises trust AI-based decision support more broadly?

Key takeaways

  • KPMG retracted AI report due to hallucinated content
  • AI models were used to analyze data but outputs weren't verified
  • Domain experts struggle to distinguish human vs. AI-generated content

Why it matters

Trust in AI-assisted knowledge work depends on human oversight. When domain experts cannot distinguish AI outputs from human work, or fail to verify them, the entire knowledge supply chain becomes questionable.