朋友们，今天给大家讲一个大型"皇帝的新衣"真人版。
事情是这样的：AI检测机构GPTZero最近给普华永道中东分部（PwC Middle East）做了个"AI体检"，查了他们2024年到2026年发布的4份研究报告。检测结果嘛……怎么说呢，有点太好笑了。
问题最严重的是一份叫《治理变革》（Transforming Governance）的报告，发表于2025年2月。GPTZero的检测显示：如果去掉参考文献等末尾内容，正文部分由AI生成的概率是100%。注意，是100%，不是99%，不是98%——是完全、彻底、百分之百的AI生成。
你以为这就完了？不，精彩的还在后面。
这份报告围绕一个叫"Citizen Pulse（公民脉搏）"的治理框架展开，说得头头是道，还声称丹麦、沙特阿拉伯、美国、澳大利亚政府都在用这个框架改善公共服务。结果呢？GPTZero查了一圈，发现"公民脉搏"这个概念，除了这份报告之外，在整个互联网上都找不到任何其他公开资料。丹麦的Borger.dk网站上没有，沙特2030愿景官网上也没有，麻省理工和世界经济论坛的文献里也没有。
也就是说，这个听起来很高大上的"公民脉搏"治理框架——是ChatGPT编出来的。
更绝的是参考文献。整份报告有17条参考文献，其中2条完全是虚构的：一篇冒充《麻省理工科技评论》的《政府服务中的人工智能》，一篇冒充世界经济论坛的《第四次工业革命：智慧城市与人工智能应用》。记者搜了一圈，这两篇文章根本不存在。
剩下的15条里，还有7条对应关系混乱、编号缺失、只链到网站首页——说白了就是凑数的。整份报告就是一个AI生成的、充斥着假概念假文献的、盖着普华永道公章的"专业研究报告"。
朋友们，你花几十万买一份四大的研报，结果拿到手的是ChatGPT一键生成的——还不如你自己花20美元充个Plus自己写呢。
不是孤例：四大集体翻车
普华永道不是第一个，也不会是最后一个。
让我们来盘点一下四大的AI幻觉翻车史：
德勤澳大利亚：2025年7月，一份价值44万澳元的政府咨询报告被查出大量虚构案例、错误引用和不存在的参考文献。其中引用了一本根本不存在的著作。最终德勤向澳大利亚政府退还了9.8万澳元。
安永加拿大：2026年5月，一份网络安全报告《攻击点：揭示忠诚度体系中的网络威胁与欺诈》因类似问题被直接撤回。
毕马威：2026年6月，一份关于代理式AI的报告被检测出45条参考文献中只有5条能准确对应真实来源。更绝的是，被列为案例的多家机构——瑞银集团、英国国民医疗服务体系NHS、瑞士联邦铁路SBB、伦敦交通局TfL——全部公开否认，说报告里关于他们使用AI的描述不准确甚至有误导性。最后毕马威把报告撤了，启动内部调查。
现在再加上普华永道中东——四大会计师事务所，一个都没跑掉。
这就很讽刺了。四大向来以专业、严谨、可信赖著称，他们的报告盖了章就有法律效力，企业上市、政府决策、投资并购都要拿他们的报告当依据。结果现在，他们的"思想领导力"白皮书，是ChatGPT写的。不仅是ChatGPT写的，还写得特别敷衍——连参考文献都懒得编圆了，直接编两个不存在的充数。
GPTZero的总结特别扎心："这些报告中错误的密集程度充分表明，它们在发布前并未经过充分人工核查。"翻译一下：写完AI生成的，连看都没看就发了。
这件事教会我们什么？
朋友们，这个翻车事件虽然很好笑，但也给我们所有人提了三个醒。
第一，不要迷信大牌机构。四大怎么样？世界顶级专业服务机构怎么样？一样用AI写报告不审稿，一样编造假概念假文献。品牌越大，不代表内容越靠谱——可能恰恰相反，因为品牌大了之后就有人"搭便车"，靠牌子吃饭而不是靠质量吃饭。
第二，看报告一定要核查参考文献。以后别人给你发一份看起来很专业的报告，先别被图表和数据吓住，翻到最后面看参考文献。如果参考文献里有你搜不到的文章、链不到具体页面的链接、格式混乱的引用——那这份报告的可信度就要打个大大的问号。AI生成的东西，最容易在参考文献这种细节上露马脚。
第三，AI时代的专业服务，正在面临信誉重构。以前"四大"两个字就是金字招牌，现在这个招牌正在被AI幻觉慢慢腐蚀。当客户发现自己花大价钱买的报告还不如自己用ChatGPT生成的好，专业服务行业的商业模式就该重新想想了。
当然，话说回来，我们也不能一竿子打翻一船人。不是所有四大的报告都是AI写的，也不是所有专业机构都在划水。但这个事件确实敲响了警钟：在AI让内容生产成本趋近于零的时代，真正有价值的不是内容本身，而是对内容质量的担保和背书。如果你连这个担保都做不好，那你的价值在哪里呢？
朋友们，下次收到一份盖着大牌机构公章的专业报告时，别急着肃然起敬——说不定它就是ChatGPT花5分钟写出来的。
今天就槽到这里，明天继续。
Friends, today we have a real-life version of "The Emperor's New Clothes" for you.
Here's what happened: AI detection firm GPTZero recently gave PwC Middle East an "AI physical," examining 4 research reports they published between 2024 and 2026. The test results… well, let's just say they're kind of hilarious.
The most problematic one was a report called Transforming Governance, published in February 2025. GPTZero's analysis showed: if you remove references and end-matter, the probability that the body text was AI-generated is 100%. Note — 100%, not 99%, not 98% — completely, totally, one-hundred-percent AI-generated.
You think that's it? No, the best part is still coming.
This entire report revolves around a governance framework called "Citizen Pulse". It talks about it very authoritatively, even claiming that the governments of Denmark, Saudi Arabia, the United States, and Australia are all using this framework to improve public services. The result? GPTZero searched far and wide and found that the "Citizen Pulse" concept doesn't appear anywhere else on the entire internet outside of this one report. It's not on Denmark's Borger.dk website, it's not on Saudi Arabia's Vision 2030 official site, it's not in MIT or World Economic Forum literature.
That is to say, this impressive-sounding "Citizen Pulse" governance framework — ChatGPT made it up.
Even better are the references. The report has 17 references total, of which 2 are completely fabricated: one posing as an MIT Technology Review article called "Artificial Intelligence in Government Services," another posing as a World Economic Forum piece called "The Fourth Industrial Revolution: Smart Cities and AI Applications." Journalists searched everywhere — these articles simply don't exist.
Of the remaining 15, another 7 have messed-up correspondences, missing numbering, or just link to website home pages — basically filler. The entire report is an AI-generated, fake-concept-and-fake-reference-filled "professional research report" bearing the PwC seal.
Friends, you spend hundreds of thousands on a Big 4 research report, and what you get is one-click generated by ChatGPT — you'd be better off spending $20 on a ChatGPT Plus subscription and writing it yourself.
Not Isolated: The Big 4 Are All Failing
PwC isn't the first, and it won't be the last.
Let's take stock of the Big 4's AI hallucination faceplant history:
Deloitte Australia: July 2025, a government consulting report worth AU$440,000 was found to contain大量 fabricated cases, incorrect citations, and non-existent references. Including a cited book that doesn't exist. Deloitte eventually refunded AU$98,000 to the Australian government.
EY Canada: May 2026, a cybersecurity report "Points of Attack: Uncovering Cyber Threats and Fraud in Loyalty Systems" was directly withdrawn for similar issues.
KPMG: June 2026, a report on agentic AI was tested — of 45 references, only 5 accurately corresponded to real sources. Even better: multiple organizations cited as case studies — UBS, NHS, Swiss Federal Railways (SBB), Transport for London (TfL) — all publicly denied the report's descriptions of their AI usage were accurate or non-misleading. KPMG eventually pulled the report and launched an internal investigation.
And now add PwC Middle East — not a single one of the Big 4 accounting firms has escaped.
This is so ironic. The Big 4 have always been known for professionalism, rigor, and trustworthiness. Their reports, sealed with their stamps, carry legal weight — corporate IPOs, government decisions, investment and M&A deals all use their reports as basis. And now, their "thought leadership" whitepapers are written by ChatGPT. Not just written by ChatGPT — written lazily, where they couldn't even be bothered to make the references look real, just fabricating two non-existent ones to pad the list.
GPTZero's summary is particularly cutting: "The density of errors in these reports strongly indicates they were not subject to adequate human review before publication." Translation: after AI generated it, they didn't even read it before publishing.
What Does This Teach Us?
Friends, while this faceplant is hilarious, it also rings three alarm bells for all of us.
First, don't worship big-brand institutions. So what if it's the Big 4? So what if it's the world's top professional services firm? They still use AI to write reports without reviewing them, still fabricate fake concepts and fake references. A bigger brand doesn't mean more reliable content — it might be precisely the opposite, because once a brand gets big enough, people start "free-riding" on it, living off the name instead of the quality.
Second, always check the references when reading reports. Next time someone sends you a report that looks very professional, don't be intimidated by the charts and data — flip to the back and check the references. If there are articles you can't find, links that don't go to specific pages, citations with messy formatting — then the credibility of that report deserves a big question mark. AI-generated content is easiest to spot in the details, like references.
Third, professional services in the AI age are facing a credibility restructuring. Before, "Big 4" was a golden brand name. Now that brand is slowly being corroded by AI hallucinations. When clients discover that reports they paid top dollar for aren't as good as what they could generate with ChatGPT themselves, the professional services industry's business model needs some rethinking.
To be fair, we shouldn't paint everyone with the same brush. Not all Big 4 reports are AI-written, and not all professional firms are phoning it in. But this incident does ring an alarm bell: in an age where AI drives content production costs toward zero, what's truly valuable isn't the content itself — it's the guarantee and endorsement of content quality. If you can't even deliver on that guarantee, then what's your value?
Friends, next time you receive a professional report stamped with a big-brand institution's seal, don't rush to be impressed — chances are, ChatGPT wrote it in 5 minutes.
That's all the roasting for today. More tomorrow.
温馨提示：1. 收到专业报告时先翻到最后看参考文献，如果有搜不到的文章、只链到首页的链接，可信度就要打问号；2. 不要迷信大牌机构的品牌背书，AI时代内容生产成本趋近于零，真正值钱的是对质量的担保——而很多机构正在丢掉这个担保；3. 自己用AI写报告/方案时，一定要人工核查事实和引用，特别是参考文献，AI特别容易在这上面翻车。
Friendly reminders: 1. When you receive a professional report, flip to the back and check the references first — if there are articles you can't find or links that only go to home pages, question the credibility; 2. Don't blindly trust big-brand institutional endorsements. In the AI age, content production costs approach zero; what's truly valuable is quality assurance — and many institutions are losing that guarantee; 3. When using AI to write reports/proposals yourself, always manually verify facts and citations, especially references — AI is especially prone to faceplants there.
PwC · Big 4 · AI-Generated · AI Hallucination · GPTZero · Citizen Pulse · Fabricated References · Professional Services · Credibility Crisis · Faceplant
