DELOITTE CHARGES $440,000 FOR CHATGPT HALLUCINATIONS, DISCOVERS HUMANS STILL HAVE TO READ THINGS

When your AI makes up fake court cases and nonexistent professors, but your invoice is very, very real

THE STORY

In October 2025, The Guardian reported that Deloitte would provide a partial refund to the Australian government after being caught using generative AI to produce a $440,000 report - a report that contained hallucinated references, made-up court cases, and citations to professors who don’t exist.

Let that sink in.

A firm that charges partners upwards of $1,000 per hour for their “expertise” and “strategic insights” outsourced the work to a language model, didn’t bother to verify the output, and submitted it to a government that was apparently paying for human intelligence.

Senator Deborah O’Neill summarized it perfectly: “Deloitte has a human intelligence problem.”

THE NUMBERS

THE ANATOMY OF A $440,000 PROMPT

Let’s reconstruct what likely happened:

Step 1: Win the Contract

Deloitte was commissioned by the Department of Employment and Workplace Relations (DEWR) to review the “targeted compliance framework” - the IT system that automates penalties for jobseekers who don’t meet their mutual obligations.

Translation: The government paid Deloitte to review the digital system that punishes poor people for paperwork errors.

Step 2: Deploy “Next-Generation Consulting Methodology”

According to the updated report’s appendix, the work “included the use of a generative artificial intelligence (AI) large language model (Azure OpenAI GPT-4o) based tool chain licensed by DEWR and hosted on DEWR’s Azure tenancy.”

Translation: They used the government’s own ChatGPT subscription to write the report the government paid them $440,000 to produce.

This is like hiring a chef for $10,000 to cook you dinner, and then watching them microwave a Hot Pocket from your own freezer.

Step 3: Skip the “Quality Assurance” Step

The AI hallucinated:

• Fake academic references from the University of Sydney and Lund University in Sweden
• A nonexistent court case called “Deanna Amato v Commonwealth” (Robodebt-related)
• Various other citations that simply do not exist

Dr. Christopher Rudge from the University of Sydney, who first spotted the errors, noted that in the “corrected” version, Deloitte didn’t just fix one fake reference - they replaced each hallucinated citation with five, six, or seven or eight real ones.

“What that suggests is that the original claim made in the body of the report wasn’t based on any one particular evidentiary source.”

Translation: They made up the claim first, then asked the AI to find sources. The AI made up sources. Nobody checked.

Step 4: Invoice

$440,000. Thank you very much.

THE CONSULTING INDUSTRY MATH

Let’s do some basic arithmetic that apparently Deloitte’s partners find too challenging:

Cost of GPT-4o via Azure API

• Input tokens: ~$0.01 per 1,000 tokens
• Output tokens: ~$0.03 per 1,000 tokens
• A 50-page report: ~75,000 tokens
• Total AI cost: ~$5

Deloitte’s Value-Add

Deloitte’s implied hourly rate: $825,000/hour

THE CORRECTION PROCESS (CHEF’S KISS)

When caught, Deloitte issued an “updated” version of the report noting “a small number of corrections to references and footnotes.”

A small number. Of entirely fabricated sources. In a $440,000 government report. About a system that punishes vulnerable people.

The department assures us: “There are no changes to the recommendations.”

Of course not. Because the conclusions were probably correct - not because Deloitte did the work, but because the problems with the compliance framework were already documented by academics, advocates, and journalists who do actual research.

Deloitte’s AI just hallucinated a bunch of fake citations to support conclusions that were already obvious to anyone paying attention.

THE “PARTIAL REFUND”

Deloitte has agreed to return the final instalment of the contract. The amount has not been disclosed.

Let’s be clear about what this means:

This is not accountability. This is a rounding error.

If you bought a car for $44,000 and discovered the engine was made of cardboard and lies, you wouldn’t accept a “partial refund.” You’d demand your money back and probably a criminal investigation.

But this is consulting. Different rules apply.

SENATOR O’NEILL’S MASTERPIECE

Labor Senator Deborah O’Neill delivered what may be the single greatest takedown of the consulting industry in parliamentary history:

“Deloitte has a human intelligence problem. This would be laughable if it wasn’t so lamentable. A partial refund looks like a partial apology for substandard work.”

And then, the coup de grace:

“Perhaps instead of a big consulting firm, procurers would be better off signing up for a ChatGPT subscription.”

The implied math:

• ChatGPT Plus: $240/year
• Deloitte report: $440,000
• Difference: $439,760
• Quality difference: Approximately zero

THE BROADER PATTERN

This isn’t an isolated incident. This is the business model.

The Big Four Consulting Playbook

1. Win contract through relationships, not capability
2. Staff with juniors who Google things and make PowerPoints
3. Leverage “proprietary methodologies” (which are just frameworks with trademarked names)
4. Deliver obvious conclusions the client already suspected
5. Bill at partner rates for associate work
6. If caught, issue partial refund and press release about “commitment to quality”

The AI Upgrade

Now add generative AI:

1. Win contract (same as before)
2. Type client’s question into ChatGPT (cost: $0.05)
3. Copy output into branded template (cost: $0 + logo)
4. Don’t verify anything (cost savings: significant)
5. Bill $440,000
6. If caught, blame the technology and issue partial refund

THE REAL SCANDAL

The real scandal isn’t that Deloitte used AI. It’s that:

1. Nobody at Deloitte Verified the Output

A $440,000 report with fabricated court cases and nonexistent professors made it through whatever passes for quality control at one of the world’s largest professional services firms.

This means either:

• a) Nobody at Deloitte read the report before submitting it
• b) They read it and didn’t notice the fake sources
• c) They noticed and didn’t care

All three options are damning.

2. The Government Didn’t Notice Either

DEWR published the report on July 4, 2025. It took the Australian Financial Review and an academic (Dr. Rudge) to spot the errors in August. The government client apparently didn’t review what they paid $440,000 for.

3. The Consequences Are Trivial

Deloitte will:

• Return some money (amount undisclosed)
• Issue corporate statements about “quality commitments”
• Continue winning government contracts next week

Nobody will be fired. No executives will be held accountable. The firm will likely use this as a case study for their “AI Ethics Framework” consulting practice.

WHAT SHOULD HAPPEN

For This Contract

1. Full refund of the entire $440,000
2. Publication of the refund amount
3. Independent audit of all Deloitte government contracts for AI use
4. Debarment period from government contracting

For the Industry

1. Mandatory disclosure of AI use in consulting deliverables
2. Verification requirements for all cited sources
3. Clawback provisions for work product that doesn’t meet professional standards
4. Personal liability for partners who sign off on fabricated work

For Governments

1. In-house capability building instead of outsourcing thinking
2. Competitive procurement that values methodology over relationships
3. Technical review panels that actually read deliverables
4. Contract structures that pay for outcomes, not hours

THE SENATOR’S PRESCRIPTION

Senator O’Neill’s suggestion wasn’t just a sick burn. It was a serious policy recommendation:

“Anyone looking to contract these firms should be asking exactly who is doing the work they are paying for, and having that expertise and no AI use verified.”

In other words: Demand to know whether you’re paying $440,000 for human expertise or $20 for an AI subscription.

Because right now, the answer might surprise you.

CONCLUSION: THE CONSULTING EMPEROR HAS NO CLOTHES (JUST A CHATGPT SUBSCRIPTION)

The Deloitte AI scandal isn’t a story about technology gone wrong. It’s a story about an industry that has been charging premium rates for commodity work for decades, and finally got caught using the commodity tools directly.

The consulting industry’s value proposition has always been: “We have expertise you don’t have.”

But if that expertise is “we know how to use ChatGPT,” then the premium disappears.

The $440,000 report contained the same information you could get by typing:

"Review the targeted compliance framework for Centrelink mutual obligations
and identify issues with traceability and system defects. Add academic
citations."

The AI added fake citations. Deloitte added a logo. The government added half a million dollars to the invoice.

And somewhere in Parramatta, a Deloitte partner is explaining to their team how “AI augmentation” is the future of consulting.

They’re right. But probably not in the way they think.

Source: The Guardian - Deloitte to pay money back to Albanese government after using AI in $440,000 report

“The future of consulting is AI-assisted expertise.” - Deloitte marketing materials, probably

“The future of consulting is paying ChatGPT rates for ChatGPT work.” - Everyone else

Disclaimer: This article was fact-checked by a human who actually read the sources. No hallucinated professors or fabricated court cases were included. Total cost of verification: approximately $0 and 20 minutes of reading. Deloitte is welcome to hire us as consultants on how to do this.