When AI breaks — who takes the blame?

CULTURE & CODE

By Joey Briones

It finally happened.

Deloitte — one of the world’s most trusted names in consulting — has had to refund the Australian government nearly USD290,000 after an artificial intelligence (AI)-generated report was found riddled with factual and analytical errors.

For a firm that sells trust, not code, the irony is impossible to ignore.

The Trust Paradox

According to reports from the Australian Financial Review, TechCrunch, and Times of India, Deloitte’s USD440,000 engagement relied on AI tools to accelerate research and drafting.

The result: efficiency without accuracy.

The company’s admission was swift, its refund voluntary, and its apology public.

But the damage was already done.

When the system that promises “intelligent assurance” produces flawed intelligence, clients don’t just question the output — they question the entire promise of responsible AI.

And Deloitte isn’t alone. Its earlier AI assurance division refund reported by Business Insider points to a broader pattern across the consulting world: firms are racing to automate their own expertise, sometimes faster than their ethics can catch up.

Automation without accountability

The incident exposes an uncomfortable truth — AI cannot yet certify its own integrity.

When algorithms audit algorithms, trust becomes recursive and fragile.

AI is brilliant at speed and pattern-making, but it lacks one crucial capacity: judgment.
That’s supposed to be the human layer — the reviewer, the consultant, the final check before “send.”

When that layer collapses under pressure for productivity, even the smartest machine becomes a liability.

The Deloitte refund is more than a financial correction; it’s a moral receipt. It shows that the cost of human absenceis measurable — sometimes down to the dollar.

The reckoning for professional services

For decades, firms like Deloitte, PwC, EY, and KPMG built empires on human intellect — the ability to reason, question, and verify.

Now they are trying to codify that judgment into algorithms.

But when the machine fails, the brand bleeds.

---

The consulting industry is entering what might be called the era of audited AI.

Every “smart” deliverable will soon need its own audit trail:Who trained it? Who checked it? Who approved it?

In other words, trust now needs metadata.

Making the changeover more human

So how do we prevent “responsible AI” from becoming another empty slogan?

1. Put people back in the loop — with purpose.

AI should augment consultants, not erase their accountability.

Every output must carry a human signature that says, “I’ve checked this, and I stand by it.”

2. Make transparency standard.

Clients deserve to know when content is machine-generated. Hidden AI use erodes confidence faster than any mistake.

3. Treat ethics as infrastructure.

Governance shouldn’t be reactive. Build moral guardrails as rigorously as you build data pipelines.

4. Reward curiosity, not compliance.

Encourage professionals to challenge AI’s conclusions. In a world of automated answers, asking better questions is the new craft.

The new definition of expertise

This moment should humble — not humiliate — the consulting world.

Because the truth is, AI didn’t fail Deloitte. Humans failed to stay human enough in the process.

The next generation of trusted advisors won’t just know how to use AI; they’ll know when not to. They’ll need to recognize that trust isn’t built on perfect algorithms, but on imperfect people who care enough to correct them.

When AI breaks, accountability shouldn’t vanish into the code.

It should remind us why judgment — ethical, human, fallible judgment — will always be the most valuable intelligence of all.