KPMG Pulls AI Usage Report After Apparent Hallucinations Surface
KPMG retracted a report on AI usage after apparent hallucinations were found in the content. Here's what it means for businesses relying on AI-generated research.
LUMIEN3 min read
KPMG has retracted a report focused on AI usage after apparent hallucinations were discovered in the content, according to TechCrunch. The withdrawal is a notable public stumble for one of the world's largest professional services firms, and it underscores a recurring problem: AI tools used to research or write about AI are not exempt from producing fabricated or inaccurate information. For any business publishing AI-assisted content, the incident is a practical warning about what can go wrong without a solid review process.
What happened
KPMG published a report covering AI usage, then pulled it after apparent hallucinations were identified in the material, as reported by TechCrunch. The firm has not, based on the available reporting, detailed exactly which claims were fabricated or how far the report circulated before it was taken down.
The story fits a familiar pattern. AI models can generate statistics, citations, and quotes that sound authoritative but have no basis in fact. When those outputs are used to build a report, especially on a fast-moving topic like AI adoption, errors can slip through review stages and reach a public audience.
Why it matters
KPMG is not a startup experimenting with new tooling. It is a major professional services firm with the resources to run proper editorial checks. If hallucinated content made it into a published report there, it signals that the problem is not just a beginner’s mistake.
There is also a specific irony worth noting. A report about AI usage was itself compromised by AI-generated inaccuracies. TechCrunch put it plainly: AI keeps proving to be an unreliable source of information about AI. That is a problem for anyone trying to use these tools to produce credible research, market analysis, or thought leadership content.
The reputational cost is real. Retracting a published report is not a quiet fix. It draws attention to the error, raises questions about internal review processes, and can undermine trust in related work the firm has published or plans to publish.
Our take
We use AI tools every day at Lumien, and we are not going to pretend they are not useful. But this story is a clear example of what happens when the output review step gets skipped or rushed, especially for anything that cites numbers, studies, or external sources.
The temptation with AI-generated research is to treat fluency as accuracy. The text reads well, the structure is logical, and the claims sound plausible. That is exactly when you need a human with domain knowledge to check the underlying facts, not just the grammar.
For smaller businesses and agencies, the risk is arguably higher. A large firm like KPMG can absorb a retraction. A small consultancy or agency that publishes a hallucinated stat in a client report or a public post has less credibility to fall back on.
A few specific checks worth building into any AI-assisted content process:
- Verify every statistic or data point against its original source, not just the AI’s summary of it.
- Treat any citation the AI produces as unconfirmed until you have pulled up the actual document.
- Have someone who knows the subject area read the draft, not just someone checking for typos.
- Be especially careful with reports about AI itself. The training data on this topic is noisy, contested, and moves fast.
What to do about it
If your team is using AI to draft reports, blog posts, or client-facing research, build a mandatory fact-check step into the workflow before anything is published. Assign a specific person to own that step. Fluent prose is not a substitute for accurate sources, and a retraction costs far more time than a proper review would have.