1 Minute AI

Deloitte Doubling Down on AI Despite Fines for AI Hallucinations

Photo by Dan V on Unsplash

Non-Medium.com users can read the full story by clicking **here**

In Australia, the government body Department of Employment and Workplace Relations hired Deloitte to write a report on a welfare compliance framework. The A$440,000 contract was assigned to Deloitte’s generative AI system (Azure OpenAI’s GPT-4o).

The resulting report, published in June, was revealed to have included false academic citations, fabricated references, and even a fictional quote from a court case. Deloitte has since promised a full refund to the government and is undertaking partial correction of errors.

Elsewhere in the world, Deloitte is investing heavily in AI. The company is teaming up with Anthropic to pilot a Claude certificate program to train 15,000 staff, and to deploy Claude to more than 470,000 employees in over 150 countries.

In this capacity, Deloitte will establish a new Centre of Excellence, create role-based AI tools for its various functions, and rely on its Trustworthy AI™ framework to mitigate issues like the ones found in the Australian report.

Taken together, these two stories show that AI presents significant opportunity — scale, efficiency, and novel capabilities — but there are risks, too. Accuracy, transparency, and verification are important. Rolling out AI at scale requires guardrails.