"Deloitte Issues Refund for Error-Ridden Australian Government Report That Used AI"

Financial Times (Ellesheva Kissin) reports:

The Big Four accountancy and consultancy firm will repay the final instalment of its government contract after conceding that some footnotes and references it contained were incorrect, Australia's Department of Employment and Workplace Relations said on Monday….

In late August the Australian Financial Review reported that the document contained multiple errors, including references and citations to non-existent reports by academics at the universities of Sydney and Lund in Sweden….

While Deloitte did not state that AI caused the mistakes in its original report, it admitted that the updated version corrected errors with citations, references, and one summary of legal proceedings….

Comments

[Stupid Government Tricks]

The cynic in me laughs at even thinking any Net Zero advocacy report could be right. "I paid for zebra shit. You sent me donkey shit. Waaahhhh!"

ETA: I may be confusing this with another report which I remember as being about Net Zero, but still an outsider's report riddled with AI hallucinations. The FT site wants me to accept cookies, so I'll have to look elsewhere.

[Stupid Government Tricks]

Yes, this report is about some welfare program unrelated to Net Zero.

[Armchair]

It gets better.

"The report reviewed departmental IT systems’ use of automated penalties in Australia's welfare system. The department said the “substance” of the report had been maintained and there were no changes to its recommendations."

So, we have AI writing a report to support the use of automated penalties in a welfare system....

"Mr. Bureaucrat, the AI gave me an incorrect fine. The system isn't working!"
"Sorry, we had an AI write a report about the AI. It's working fine".

[Brett Bellmore]

They seem to be getting off light, frankly.

[Armchair]

https://techcrunch.com/2025/10/06/deloitte-goes-all-in-on-ai-despite-having-to-issue-a-hefty-refund-for-use-of-ai/

Deloitte announced Monday plans roll out Anthropic’s chatbot Claude to its nearly 500,000 global employees on Monday. Deloitte and Anthropic, which formed a partnership last year, plan to create compliance products and features for regulated industries including financial services, healthcare and public services, according to an Anthropic blog post. Deloitte also plans to create different AI agent “personas” to represent the different departments within the company including accountants and software developers, according to reporting from CNBC."

Too bad we can't short sell the stock.

[Armchair]

I do have to ask though...how do companies think this is possibly a good idea?

The issues with LLM are well known. They will tell you what you wants to hear, and are more than willing to make up references, quotes, citations, etc to support the case. I've experimented with them, and they will just tell you wrong information, in a definitive tone.

If this was an individual doing it, the companies would fire them instantly. If the individual said "yeah, I couldn't find the reference or quote to support this, so I just made it up. It was more efficient and faster to do that."...the company would fire the person.

If there was a contractor that offered the company the capability to write reports or case briefs quickly and cheaply* (* but there's a reasonable chance that we'll just fabricate key references and quotes for efficiency)....a company wouldn't use them. It would be a automatic disqualifier.

But do the same thing, and call it "AI"...and suddenly they're on board?

[Stupid Government Tricks]

I think it's the same reason I commented on for some lawyer AI misuse. The highly paid experts have a lot of human assistants doing all the scut work, then the experts did some final editing and summarizing and signed off on it. Human assistants aren't likely to lie and hallucinate and take short cuts because they'd get fired.

Then AI seduces the experts with the promise of being faster and cheaper, along with not being humans who take sick days and vacations and don't work 24 hours a day. The output looks reasonable, all nicely formatted, no typoes ... and the experts still believe assistants don't like and make stuff up.

[Brett Bellmore]

And maybe a year or two ago there could have been something vaguely like an excuse for that. By now? EVERYBODY knows you can't trust AI output.

I'd be more likely to believe that some of guy on the scut work detail who figures they're on their way out the door soon anyway decided to retire in place, and have the AI do the hard work until the hammer actually fell.

[Armchair]

I'm not sure Brett....

I think the problem is that the leaders "know" what looks good. If they got a piece of crap writing from a 1st year associate full of typos and editing errors, they'd assume the references were wrong as well. And they'd send it back.

But if the leaders see something that "looks good" (ie, no typos, good writing style, etc), they assume the rest of it is correct. If someone took the care to go through and edit the document for typos and writing, of course they'd also do the references correctly. And, largely speaking, if the writer was human...they'd be correct. But AI is different.

AI "looks" good...but the hard work to put into the document, it skips over (it's not really capable of it). Editing, spell check...looks good. So, leaders believe it. It's fast, tells them what they want to know and hear, looks pretty.

[Brett Bellmore]

Sure, and that's where my "a year or two ago there could have been something vaguely like an excuse for that" comes in.

Anybody who doesn't know at this point that LLM's produce well formatted results that are often simply fiction hasn't done the due diligence necessary to be using them. So there may be explanations, but no excuses.

I simply don't see how anybody in the legal community could plausibly not know at this point that AIs hallucinate, and that if you use AI generated content you need to have a human fact check everything. Even if the guy at the top slept through the last two years, his underlings would know and mention it.

So I asked, "Who might know that, and not care?

[SRG2]

As an ex-Deloittee, I found this hilarious

[James K. Polk]

Touche would never have used AI.

[Bored Lawyer]

Exhibit 1006 that AI produces garbage.

Query: In 20 years, will we view the hype about the AI the same as we today view the dot com bubble?