KPMG Report on AI Safety Undermined by AI Errors

Researchers found that a KPMG document advising businesses on AI safety contained dozens of made-up sources—fake author names, fake dates, fake journals. Of 45 citations, only 5 were real. These are typical errors from AI systems that confidently generate plausible-sounding but entirely false information. The catch: the report was supposed to teach companies how to avoid exactly this kind of AI failure.