Image by BoliviaInteligente on Unsplash
Turns out, doubling down on AI doesn’t mean you won’t trip over it. On the same day Deloitte proudly announced a major enterprise deal with AI startup Anthropic, it had to issue a refund for a government-commissioned report loaded with AI-generated errors. Yep—AI giveth, and AI taketh away.
Let’s break it down.
A Pricey Mistake in Australia
Earlier this year, Deloitte prepared what was supposed to be an “independent assurance review” for Australia’s Department of Employment and Workplace Relations. The cost? A hefty A$439,000.
But there was a problem. (Well, several.)
The Australian Financial Review revealed in August that the report contained numerous errors, including citations to academic sources that didn’t exist. In other words, AI-generated “hallucinations” made their way into official government documentation. Talk about awkward.
Image by Marija Zaric on Unsplash
The department has since published a corrected version of the report on its website. Deloitte is now repaying the final installment of that contract. Meanwhile, questions loom about the consulting giant’s internal review processes and its ability to responsibly use AI tools.
Meanwhile, a Major AI Deal
Now for the plot twist: On the very same day this refund news broke, Deloitte went public with a massive AI move.
The company is partnering with Anthropic, the creators of the Claude chatbot, in what is being called Anthropic’s largest enterprise deployment to date. Deloitte says it plans to roll out Claude to nearly 500,000 employees across its global network.
That’s not just a few office pilots—this is full-on integration.
The goal? To build AI tools tailored for regulated industries like finance, healthcare, and public services. Deloitte also wants to create internal “AI personas” that reflect different job roles, like accountants and software engineers, to streamline its workflows.
In a blog post, Ranjit Bawa, Deloitte’s Global Technology and Ecosystems and Alliances Leader, framed the partnership as a big bet on the future of responsible AI:
“Deloitte is making this significant investment in Anthropic’s AI platform because our approach to responsible AI is very aligned, and together we can reshape how enterprises operate over the next decade.”
Sounds confident—but clearly, the road to “responsible AI” has some potholes.
Image by Dhaya Eddine Bentaleb on Unsplash
AI Growing Pains Aren’t Unique
Deloitte’s situation isn’t an outlier. Other companies are grappling with the same issues:
- Earlier this year, Anthropic’s own legal team had to apologize after Claude cited fictional legal precedents in a music publishing dispute.
- Amazon’s enterprise AI tool, Q Business, struggled with basic accuracy in its rollout, according to internal findings.
- Even the Chicago Sun-Times ran a summer reading list that included fake books generated by AI.
These mishaps highlight a key tension: AI is powerful, but still deeply flawed. Especially when used without rigorous human oversight.
What This Means for the Future
Here’s the takeaway: even companies at the forefront of tech innovation are still figuring out how to use AI without falling on their face.
Deloitte’s new alliance with Anthropic is a bold move—and could lead to some genuinely useful applications for AI in large-scale corporate settings. But it also serves as a loud reminder: AI isn’t plug-and-play. It needs careful handling, especially when used in sensitive, high-stakes contexts like government reports or public health tools.
The irony of Deloitte’s refund arriving the same day as its AI partnership news isn’t lost on anyone. Maybe it’s a coincidence. Or maybe it’s just 2025 being 2025.
Either way, the message is clear: investing in AI is easy. Getting it right? That’s the hard part.
Image by Galih Ass on Unsplash
Keywords: Deloitte AI deal, Anthropic Claude, AI hallucinations, Deloitte refund, enterprise AI solutions, responsible AI, AI misuse risks, Australia AI report, corporate AI tools, Generative AI in consulting
