Deloitte’s A$440,000 ‘Human Intelligence Problem’ 

On Monday, Deloitte made two contrasting headlines. One, it announced a major AI deal with Anthropic to roll out Claude across its global workforce of nearly half a million people. Two, it agreed to issue a refund to the Australian government for an AI-assisted report filled with fabricated references and hallucinated citations which was published in July.

Both the developments were reported the same day.

On paper, the partnership with Anthropic marks Deloitte’s biggest AI push yet. But the refund story undercuts it with perfect irony. As the company pushes deeper into AI, it is also being reminded of why AI still needs human supervision.

The Australian Department of Employment and Workplace Relations confirmed that Deloitte’s A$439,000 (US$290,000) “independent assurance review” contained fake citations and inaccurate references. The firm will return part of the payment. The errors were traced back to the use of an Azure OpenAI system using the GPT-4o model. 

Despite the errors and admitting the problems, Deloitte said that the use of AI did not alter the “substantive content, findings or recommendations” of the report. “The updates made in no way impact or affect the substantive content, findings and recommendations,” the firm said in the amended version.

But lawmakers and experts seized on the incident as a warning. Australian Sen. Deborah O’Neill at a senate hearing blasted that “Deloitte has a human intelligence problem.” She added that “this would be laughable if it wasn’t so lamentable. A partial refund looks like a partial apology for substandard work.”

She quipped that government agencies might be “better off signing up for a ChatGPT subscription” than rely blindly on big consultancies, The Guardian reported.

The Oversight, Not a Scandal

But Gaurav Vasu, founder of UnearthInsight, calls it an oversight. “I don’t see a significant impact of this, to be honest,” he told AIM. “Every consulting firm is using AI tools. McKinsey is big on in-house proprietary solutions with OpenAI. This is not a scandal, it’s a slip.”

Vasu believes that such errors are inevitable at this stage. “It’s very difficult to stop anyone from using AI now. It’s like Google. You can find wrong information there too. AI tools are no different.”

He pointed out that this isn’t new to consulting or content industries. “Even in traditional setups, people have cited wrong sources or used bad data. The difference now is that AI is faster — it amplifies both accuracy and error.”

Deloitte has maintained that the inaccuracies didn’t affect the report’s recommendations, which the government accepted. “They admitted there were errors not source-validated, but the findings still stand,” said Vasu. “It’s better to have small mistakes now than a massive one later. These incidents help everyone figure out how to validate data better.”

If anything, he said, this episode will nudge consulting firms to be more careful. “Governments will now validate information more strictly, but that doesn’t mean they’ll stop using consultants or AI. They’ll just keep a closer human eye on it.”

Why It Won’t Slow Down AI

Despite the timing, Deloitte’s new alliance with Anthropic signals the opposite of retreat. The firm is doubling down on AI. It plans to deploy Anthropic’s Claude chatbot across its global workforce and co-develop AI compliance tools for regulated sectors like healthcare and financial services.

Vasu said the episode doesn’t change much. “AI in consulting is too entrenched now. Nobody’s rolling back billions in investments because of one faulty report.”

The bigger takeaway, he said, is about accountability. “The assumption everywhere — from governments to clients — is that there’s a human in the loop. AI doesn’t hand reports directly to the government. A human does. That part has to get stronger.”

The “human in the loop” has become the industry’s safety valve. Every consulting firm says it has one. Yet the Deloitte case shows how that loop sometimes breaks. “It just means they’ll tighten their review processes. That’s all. It’s not like thousands of analysts were fired when Azure OpenAI was brought in,” Vasu said.

AI’s Growing Pains

For consulting firms, AI hallucinations are the new spreadsheet errors. Embarrassing but not existential. And like any new technology, they’re being ironed out as the systems evolve.

This incident also underlines the strange balance between consulting firms and the AI companies they depend on. Deloitte’s deal with Anthropic mirrors what Accenture, PwC, and EY are already doing with OpenAI and Google Cloud.

Accenture has committed $3 billion to generative AI by 2030. EY plans to invest $1.4 billion in the next five years. Deloitte is increasing its AI investments by 30% in FY26. None of them are slowing down.

Read: Why Are Consulting Firms Building LLMs

Because while AI startups build tools, consulting firms sell transformation.

Vasu said that even if startups build their own consulting arms later, it won’t be overnight. “The day OpenAI or Anthropic start hiring 100,000 people and competing with Deloitte or Accenture, that’ll be a different story. For now, they need each other.”

AI hallucinations will happen again. The same way typos, bad charts, and wrong numbers have always figured in consulting. The difference is visibility. Every AI error now becomes a headline.

But for a firm like Deloitte, that’s the cost of being an early mover. It is already embedding Claude, working with Google Cloud on AI agents, and co-developing interoperability protocols with ServiceNow.

The Australian refund, for all its headlines, is a footnote in that story. “Nobody’s shutting the door on AI. In fact, this will accelerate better checks and better tools. And it’s good that it happened now, with a small fine, not later with a big one,” Vasu said. 

The post Deloitte’s A$440,000 ‘Human Intelligence Problem’  appeared first on Analytics India Magazine.