⚖️ Productivity or rigor? AI just reminded Deloitte that the law doesn’t hallucinate.
The Australian government commissioned Deloitte to audit a social welfare program.
Contract amount: €247,000.
Objective: to identify flaws in a system accused of unfairly penalizing job seekers.
But in September 2025, authorities discovered that the report delivered by Deloitte contained… sources fabricated by AI academic references attributed to real researchers, but which never existed.
The government then demanded a partial refund of the contract, citing a serious breach of diligence expected from an international consulting firm.
💬 The facts behind the failure
Deloitte reportedly used a GPT model hosted on Azure OpenAI to assist in drafting the report.
The firm claimed that “the errors did not affect the substance of the conclusions.”
But can conclusions truly be reliable when their very foundations are false?
This case highlights a classic pitfall: blind trust in the machine.
AI didn’t “invent” maliciously, it did what it’s designed to do: produce plausible text.
It is up to the human, and only the human, to verify the truthfulness of the output.
🔍 A reminder for legal practitioners
This case reinforces an obvious principle 👉 Delegation does not eliminate responsibility.
Whether it’s a public audit, a contract analysis, or a legal memorandum, human diligence remains a non-negotiable pillar.
Law, whether civil, administrative, or ethical, is built on the idea that professionals must perform their duties with competence and vigilance.
The tool neither erases the fault nor the duty to verify.
An incorrect citation, a biased argument, or an unrevised draft each can engage the professional’s liability, not the machine’s.
🧠 The broader lesson
This fiasco goes beyond Deloitte.
It raises a deeper question about AI’s place in knowledge-based professions.
Can we delegate critical thinking, verification, or judgment? Certainly not.
Even the upcoming EU AI Act mandates effective human oversight (“human-in-the-loop”) for high-risk systems, particularly those used in public decision-making.
AI can accelerate, structure, assist, but it does not understand.
Law, on the other hand, rests on understanding, meaning, and nuance.
The Deloitte case isn’t a scandal, it’s a professional ethics reminder: When productivity takes precedence over rigor, credibility collapses.
