Friends, today's magical realist AI industry drama stars PricewaterhouseCoopers.

Here's what happened: someone discovered that four professional research reports published by PwC Middle East between 2024 and 2026 were, upon inspection, full of AI hallucinations — fabricated concepts, erroneous data citations, convincing-looking made-up references, unclickable fake footnotes, misattributed viewpoints. The most absurd case was a report called "Transformation Governance" — run through AI detection tools, the results showed: the main body was 100% AI-generated, not a single word written by a human. They didn't even edit it. They exported ChatGPT's output directly to PDF, stamped it with the PwC seal, and sent it out as a "professional research report" to clients.

Friends, this is like going to a Michelin three-star restaurant where the menu says "chef's special truffle wagyu," and the kitchen serves you a heated-up ready meal with a fake origin label stuck on the package — imported French truffles, but actually mushroom sauce made in Henan.

And this wasn't an intern accidentally sending the wrong file. This was four reports spanning two years, covering cutting-edge topics like AI and electric vehicles. That means for two full years, the entire PwC Middle East team — from the consultants writing the reports to the managers reviewing them to the partners signing off — none of them noticed, or more precisely none of them cared, that these reports were made up by AI.

AI Hallucinations Just Smashed Professional Services' Gold Standard

Why is this particularly ironic? Because who is PwC? It's the largest of the Big Four accounting firms, synonymous with "professionalism," "rigor," "trustworthiness." Public companies need their audit to release earnings; cross-border M&A needs their due diligence to sign; companies pay millions and tens of millions for their consulting, buying exactly three things: credibility.

Now you're telling me the reports they publish under their official seal have references that were completely made up?

Let's look at some of the absurd content in these reports: one report mentioned an "industry-standard AI governance framework" with a very impressive-sounding name, but searching the entire internet, no such framework exists — AI made it up; one report cited a "2025 McKinsey Global AI Research Report" — McKinsey did publish a similar report, but the data cited in the PwC report didn't match the original at all; AI just made up numbers. Another report listed over a dozen references where clicking the links either gave a 404, led to unrelated papers, or even had fabricated author names.

The funniest part was a fabricated "expert viewpoint" claiming "Professor Smith at Harvard Business School's 2025 research shows that…" — someone checked HBS's faculty roster for the past five years and there wasn't a single Professor Smith doing related research. AI didn't just make up the research conclusion; it made up the person too.

Three Wake-Up Calls for Everyone Using AI

PwC's mishap isn't an isolated case; it's an inevitable mess as AI becomes ubiquitous. It's just that this time the crash happened at the professional institution that least should have crashed, which makes it particularly instructive. Three reminders for everyone writing things with AI:

First, in professional content domains, AI generation absolutely requires 100% human review. AI is an excellent writing assistant — it can help outline, brainstorm, draft — but any professional content going external — reports, whitepapers, research notes, client deliverables — must be reviewed word-by-word by a competent human. AI confidently fabricates facts, cites nonexistent references, and gives incorrect data; this isn't a bug, it's a fundamental property of current large models. PwC's problem this time wasn't "using AI" — it was "using AI and then hitting publish without even reading it."

Second, don't blindly trust output from big-name institutions.Many people see the PwC, McKinsey, or BCG logo and assume the content must be solid. That's no longer true in this era. Big-name institutions are still staffed by people, and people cut corners, get complacent, and take shortcuts to meet deadlines. Going forward, when you see any institutional report, especially on fast-moving topics like AI or clean energy, keep a critical eye — cross-verify key data points yourself; click on references to check if they actually exist.

Third, in the AI era, "being careful" becomes even more valuable. When everyone is using AI to one-click-generate content, people willing to spend time verifying facts, checking data, refining arguments, and taking responsibility for every word they produce become more scarce and more valuable. AI lowered the barrier to content production, but it didn't lower the bar for "good content." True professionalism isn't about what tool you write with; it's about whether you have respect for what you write.

Reportedly, PwC Middle East has launched an internal investigation, and the relevant team leaders may face accountability. But the reputational damage done here can't be fixed by firing a few people. After all, when clients hire you for professional services, they're paying for your judgment and responsibility, not for how fast you can press the "generate" button.

That's all the roasting for today. More tomorrow.