KPMG: 56% of AI users made costly mistakes from unverified outputs
In brief
- KPMG surveyed 48,000 people across 47 countries on AI trust and adoption in 2025.
- 73% reported personal benefits from AI, primarily efficiency gains.
- 56% made costly mistakes by relying on unverified AI outputs.
- EY retracted a cybersecurity report after discovering 16 of 27 citations were AI-fabricated.
- Professional services firms invest heavily in AI governance frameworks.
The Trust Gap
KPMG's 2025 global study on trust in artificial intelligence surveyed over 48,000 people across 47 countries, conducted in collaboration with the University of Melbourne. The findings expose a critical vulnerability in how organizations deploy generative AI: enthusiasm for productivity outpaces caution about accuracy.
The numbers tell the story. 73% of respondents said they've observed personal benefits from AI, primarily around efficiency gains. That confidence drives adoption. But 56% of those surveyed reported making mistakes because they relied on unverified AI outputs. The gap between perceived benefit and actual risk is where organizations stumble.
When AI Lies with Confidence
KPMG's own study explicitly identifies hallucinations as a major limitation of generative AI. Hallucinations—when AI confidently presents completely fabricated information as fact—have already cost real money at major firms. EY had to retract a cybersecurity report in May 2026 after discovering that 16 of its 27 citations were completely fabricated by AI. Deloitte agreed to refund portions of a government contract after AI hallucinations were found in one of its reports.
These aren't edge cases. They're warnings.
Building Guardrails
The solution isn't to abandon AI. It's to govern it. Professional services firms are investing heavily in AI governance frameworks to capture AI's efficiency benefits while building human oversight to catch hallucinations. That means human review before outputs leave the organization, citation verification, and audit trails for any AI-generated work product.
The KPMG study shows that AI's productivity edge is real. But so is the cost of trusting it blindly. Organizations that move fast without verification will find themselves in the same position as EY and Deloitte—explaining why their AI made things up.
