AI News & Trends

Deloitte charges €250k for a report full of AI hallucinations: welcome to the era of “workslop”

4 MIN
October 16, 2025

Deloitte Australia has just experienced a nightmare: 250,000 euros invoiced for a government report full of errors generated by ChatGPT. Fictional quotations, invented case law, non-existent references... This case reveals the “workslop”, this fake AI job that affects 40% of American companies.

The case: 237 pages of hallucinations billed

The Australian government commissions Deloitte to produce an official report on its social assistance system. Budget: 250,000 euros.

The final deliverable contains:

  • Fictional quotes attributed to real experts
  • Bibliographical references do not exist
  • Court decisions made up from scratch
  • Wrong factual data

A single researcher is enough to identify the deception. Deloitte must acknowledge the unverified use of ChatGPT, publish a corrected version and reimburse 56,000 euros.

Beyond the financial damage, it is a major blow for a firm that sells precisely... reliability.

The workslop: when AI produces fake good work

The “workslop” refers to this AI production that seems professional on the surface but proves to be unusable on the inside.

The alarming numbers (BetterUp Labs & Stanford):

  • 40% of US employees confronted with workslop
  • 2 hours lost per task correcting AI work
  • 95% of businesses without measurable ROI from their AI tools
  • Confidence in free fall in teams

Typical symptoms: Volumetric but empty content, unverifiable sources, too smooth style, multiple factual errors.

Why AI hallucinations are more dangerous

AI errors are radically different from human errors:

  • They seem professional : flawless structuring, appropriate technical vocabulary
  • Absolute trust : no warning signal, no hesitation
  • Massive ladder : can contaminate 237 pages without fatigue
  • Unforeseeable : variable results for the same prompt

Our brains detect human hesitation. Faced with the assurance of an AI, our usual safeguards are deactivated.

The illusion of the magic co-pilot

“I didn't have time to check” became the recurring excuse. But this logic is absurd: if you don't have time to check, you don't have time to use AI.

The time “saved” in generations is lost (and more) in correcting or managing crises.

With AI, yes. Through AI, no.

AI should be a tool at the service of humans, not a substitute.

Smart use (what Deloitte should have done):

  • Structuring an initial plan
  • Generate hypotheses to be tested
  • Rephrase passages
  • Accelerate fitness

Risky use (what they did):Delegate the production of factual content without verification.

How to integrate AI intelligently

Training in critical use, not just technique

At HEYIA Studio, we train our customers to:

  • Identify relevant vs. at-risk use cases
  • Check systematically (checklist, sources, consistency)
  • Understanding the limitations and failure modes of AI
  • Maintain critical thinking and final judgment

The 4 pillars of an effective framework

1. Clear guidelinesAuthorized vs prohibited uses, validation levels, defined responsibilities.

2. Continuing educationReal technical understanding, business use cases, practical exercises.

3. Verification processSpecific checklist, double validation on critical content, traceability.

4. Culture of responsibilityAI does not disempower, the human signature is always binding.

Questions to ask yourself before deploying

  • Do your employees understand how AI hallucinates?
  • Do you have appropriate verification processes?
  • Is the final responsibility still assigned to a human?
  • Do you measure the real ROI (time saved - correction time)?

If you answer “no” to multiple questions, you're not ready.

Conclusion: AI requires more human expertise, not less

The Deloitte case illustrates a dangerous illusion: that AI can replace human intelligence and rigor.

The essential lessons:

  • AI is hallucinating by nature, it is not a correctable bug
  • Workslop already affects 40% of organizations
  • “With AI” is very different from “through AI”
  • Using AI properly requires more expertise, not less

Ironically, it is necessary to understand the domain to detect hallucinations, to have perspective to validate the relevance, to maintain the requirement in order not to be seduced by volumetric but empty content.

Deloitte's 250,000 euros are the price of a lesson that all organizations should learn without having to pay for it themselves.

And you, how do you supervise the use of AI in your organization?

Article written by
Benjamin BENOLIEL
Co-founder & Head of Sales

Each image tells 
a story. Ready to create your own?

Let's discuss your ideas, we'll take care of bringing them to life.