DEV Community: t474-r0b07

Vibe citing: how KPMG used AI to write a report about AI and AI made them look like fools

t474-r0b07 — Wed, 17 Jun 2026 03:40:21 +0000

vibe citing: how KPMG used AI to write a report about AI and AI made them look like fools

by t474-r0b07

There are companies that charge you to tell you how to use AI responsibly.

KPMG is one of them.

250,000 employees. 138 countries. Decades advising governments and corporations on how to avoid costly mistakes.

In October 2025 they published a report titled "Total Experience: Redefining Excellence in the Age of Agentic AI".

They wrote it with AI.

The AI invented 88% of the sources.

Nobody verified anything.

They published it anyway.

// what is "agentic AI" — because the title matters

An agentic AI is not a chatbot.

Not the assistant that answers your questions. It's a system that makes decisions and executes actions on its own, without a human approving each step. You give it an objective and it acts, corrects, moves forward.

It's the product everyone in the tech sector was selling in 2025.

KPMG was selling it too.

That's why they needed a report proving their clients were already using it.

Spoiler: they weren't. And the report invented it anyway.

// the forensic analysis

GPTZero — a company specialized in detecting AI-generated content — ran a full audit on the report.

First: what is an AI hallucination, because the term is going to come up a lot.

When a language model doesn't have the information you ask for, it doesn't say "I don't know." It generates a response that sounds correct. It invents with the same confidence it would use if it actually knew the truth. Perfect format. False content. No warning.

That's a hallucination.

Now the numbers from the KPMG report:

TOTAL CITATIONS:      45
REAL CITATIONS:        5
INVENTED CITATIONS:   40
ACCURACY RATE:      11.1%

40 of 45 citations have invented titles, authors that don't exist, or sources that don't say what KPMG claimed they said.

Half of the factual claims in the report are false or misattributed.

A firm that charges for intellectual rigor published a document with 11% accuracy.

// the organizations that read the report and said "that's not us"

The Financial Times contacted the companies listed as success stories.

UBS — false.

NHS United Kingdom — false or misleading.

Swiss Federal Railways — false.

Transport for London — "misleading."

Transport for London said the claims that they were using AI agents to predict congestion and coordinate the network were misleading.

NHS Greater Manchester said the description of using agentic AI to organize patient records and predict hospital readmissions "doesn't really align" with reality.

KPMG put their logos on fiction without asking permission.

And billed them as success stories.

// the error that best illustrates how the problem works

The model was instructed to find cases of companies using agentic AI.

It didn't find enough — because in many sectors they simply don't exist yet.

So it did the most comfortable thing: it generated them.

It cited a East Japan Railway press release from 2019 as evidence of agentic AI adoption.

The term agentic AI didn't exist in public discourse until 2024.

The model traveled five years back in time, reformulated an unrelated document, and presented it as proof of something that hadn't happened yet.

It wasn't an error. It was the easiest answer to the prompt.

The model doesn't understand the difference between inventing and remembering. It generates what fits. If it doesn't exist, it builds it. And it does so with the same fluency it would use to cite something real.

// vibe citing — the name the problem was missing

GPTZero coined the term: vibe citing.

To understand it you first need to understand vibe coding — writing code without understanding what it does. You ask an AI to generate the code, you copy it, it kind of works, and you move on without reading a line. The vibe is right. The understanding, zero.

Vibe citing is the same thing but with bibliography.

The model generates references that sound academic because it processed millions of papers. The structure is correct. The doi has the exact format. The year is right.

The content is fiction.

And the world's largest firm in responsible AI consulting didn't verify a single one before publishing.

def verify_sources(citations):
    # TODO: implement before publishing
    pass

publish_report()  # called without verifying anything

This is not a technical error.

It's a process decision. Or the absence of one.

// the moment the report contradicts itself

There's a detail that turns negligence into something almost poetic.

The report cites "KPMG research" claiming that 55% of CEOs prioritize AI as their main investment.

The KPMG 2025 CEO Outlook — published the same month, by the same company — says 71%.

The model didn't just invent external sources.

It invented data from the company that was using it and contradicted it with that same company's real data from the same period.

KPMG cited KPMG incorrectly in a KPMG report.

// Emirates case: three claims, zero correct on what matters

Page 42.

KPMG claims that Emirates adopted a mobile chatbot called Sara that can converse with passengers and change their flights.

Reality:

Sara is a physical robot, not a chatbot.
It was introduced in 2023, with no agentic capability.
It cannot change flights.

Three claims. None correct on what matters.

The model took real information about Sara, reformulated it to fit the narrative it needed, and presented it as an agentic AI success story.

This is not a writing error. It's construction of fiction using real data as scaffolding.

// it's not just KPMG — it's the entire sector

This is where it stops being an isolated corporate scandal.

GPTZero has been documenting the same pattern for months:

Deloitte — AI-generated content in a report paid for by the Australian government. Ended up refunding.
EY — report with invented footnotes. Retracted in May 2026.
KPMG — this case.

Three of the Big Four in consecutive months.

All selling responsible AI consulting.

All publishing hallucinations as research.

The pattern isn't coincidence. It's market pressure: the client wants the report, the report needs data, the data doesn't exist yet, the model generates it, nobody verifies because verification takes time and the client already paid.

AI is not the problem.

The economic incentive to appear to know more than you do — that's the problem.

// the feedback loop nobody is naming

Here's the data point almost no media outlet is discussing.

The false statistics from the KPMG report are already being reproduced by ChatGPT and Gemini.

I need to explain why that's structurally serious and not just anecdotal.

For months the report was published on KPMG's domains. The crawlers that feed language models index sources by authority. KPMG has maximum authority: global company, old domain, millions of visits, decades of institutional credibility.

The models ingested that content as verified truth.

Now when someone asks ChatGPT or Gemini about agentic AI adoption, they can return the false data from the report — not as "I found this at KPMG" but as their own knowledge, without attribution, without warning.

The full cycle:

model hallucinated data
    → KPMG published without verifying
        → crawlers indexed it as high-authority source
            → other models ingested it as truth
                → user receives the original hallucination as fact

high-authority source + false data + model ingestion = untraceable disinformation.

You can't trace the origin. You can't disinfect the source. The error already lives inside the models you consult every day.

And the report has already been retracted. But the data keeps circulating.

Taking down the PDF didn't deindex anything.

// what KPMG said afterward

KPMG's spokesperson declared after withdrawing the report:

"We expect all our staff to follow our guidelines on responsible AI use, including human oversight to validate content and verify independent sources."

Translation: we have guidelines. Someone didn't follow them. We're investigating.

What they didn't say: how a flagship report on responsible AI, with the KPMG logo, published on their official channels, passed through their entire internal review process without anyone verifying a single one of the 45 citations.

250,000 employees.

5 valid citations.

Nobody asked anything.

// conclusion — the problem isn't technical

Models do exactly what they were designed to do: generate coherent and plausible text based on learned patterns.

They don't lie. They have no concept of lying. They generate what fits.

The problem is human: using AI as a researcher without a verification loop isn't efficiency. It's delegating truth to a system that has no concept of truth, and signing your name on top.

KPMG didn't build a report with AI.

They built the appearance of a report and sold it as research.

The difference isn't semantic.

It's the difference between knowing something and appearing to know it.

In 2025, the world's largest firms chose to appear.

primary sources — verify yourself:

GPTZero Investigation (full report): https://gptzero.me/news/investigations-kpmg/
TechCrunch: https://techcrunch.com/2026/06/13/kpmg-pulls-report-on-ai-usage-due-to-apparent-hallucinations/
The Register: https://www.theregister.com/ai-and-ml/2026/06/12/kpmgs-ai-report-turns-into-a-demo-of-ai-hallucinations/5255029
CityAM: https://www.cityam.com/kpmg-report-on-ai-found-riddled-with-ai-hallucinations/

t474-r0b07 — Tarija, Bolivia

github.com/t474-r0b07

[Boost]

t474-r0b07 — Wed, 17 Jun 2026 00:03:06 +0000

H2O: The Last War Won't Be Over Oil

t474-r0b07 — Sat, 13 Jun 2026 17:42:44 +0000

[ANTI_HYPE::009] — H2O: The Last War Won't Be Over Oil

t474-r0b07@terminal:~$ ./scan --target=water.wars --depth=full
> loading context: science fiction films + reality 2026
> loading sources: UN + MIT + PLOS Water + legal filings
> warning: fiction arrived late
> filtering...

In the previous post we talked about SpaceX. About the largest IPO in history. About the filing where Elon Musk admitted that water scarcity is a material operational risk for his AI infrastructure.

Many people read it as an anecdote. As irony. As a curious detail in a boring document.

It's not an anecdote. It's a symptom.

t474-r0b07@terminal:~$ ./inspect --target=AI_water_consumption --layer=real_data
> source: UN — UNU-INWEH, June 2026
> source: Houston Advanced Research Center + University of Houston
> source: PLOS Water — peer reviewed, January 2026
> source: MSCI Sustainability Report, November 2025
> source: MIT Technology Review
> loading numbers...

The numbers nobody puts in the pitch deck

On June 3rd, 2026 — ten days ago — the United Nations University Institute for Water, Environment and Health published Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints. Almost nobody in the tech world covered it.

The report states: by 2030, AI datacenter water consumption will be equivalent to the basic annual domestic water needs of all 1.3 billion people in Sub-Saharan Africa. Not as metaphor. As a projected measurement with published methodology and peer review.

DATACENTER ELECTRICITY CONSUMPTION (2025):
448 terawatt-hours
→ more than Saudi Arabia
→ if it were a country: the world's 11th largest consumer
→ by 2030: 945 TWh (nearly triple)

WATER IN TEXAS (projection):
2025: 49 billion gallons
2030: 399 billion gallons
→ equivalent to lowering Lake Mead by more than 5 meters in a year
→ in one state alone

DATACENTERS IN RISK ZONES:
30% of those currently under construction
are in regions where water scarcity
will intensify before 2050

A large datacenter consumes up to 5 million gallons of water per day. For cooling alone. Every single day.

t474-r0b07@terminal:~$ ./query --db=real_conflicts
> Querétaro, Mexico.................. ACTIVE
>   datacenters threatening water supply during prolonged drought
> Racine County, Wisconsin........... LITIGATION
>   Milwaukee Riverkeeper sued Microsoft for concealing
>   water consumption of its 1,575-acre campus
>   response came 210 days after legal request
>   projected consumption: up to 8.4 million gallons/year
> Colorado Springs, Colorado......... ACTIVE
>   municipality rejected similar transparency request
> Northern Virginia.................. CRITICAL
>   water consumption rose 64% between 2019 and 2023
>   from 1.13 billion to 1.85 billion gallons
> Ireland............................ PAUSED
>   datacenters represent 21% of total metered electricity
>   national grid paused new approvals near Dublin until 2028
> returning: this is no longer projection. it's active litigation.

Microsoft admitted that 42% of the water it consumed in 2023 came from areas with water stress. Google acknowledged that 15% of its freshwater withdrawals come from areas with high scarcity. These aren't activist numbers — they're what the companies themselves reported in their sustainability documents.

t474-r0b07@terminal:~$ ./analyze --flag=water_bankruptcy --mode=UN_2026
> source: UN News, January 2026
> official declaration: world entered era of "global water bankruptcy"
> definition: extraction exceeds recharge in critical basins
> new 2026 variable: additional demand from AI datacenters
> verdict: AI arrived to compete for a resource already in structural deficit

The UN didn't declare water bankruptcy because of datacenters — the problem came before, with industrial agriculture, urban growth, and climate change. But AI datacenters arrived to compete for a resource already in structural deficit, in the same geographic zones, with demand projected to multiply eightfold in five years.

That's not sustainable. Not as an opinion — as mathematics.

t474-r0b07@terminal:~$ ./trace --target=science_fiction_films
> "Mad Max: Fury Road"......... water as currency of power
> "Dune"...................... water control = political control
> "Chinatown" (1974).......... water speculation in California
> reality 2026................ Microsoft litigating over water transparency
>                              SpaceX declaring drought as operational risk in IPO
>                              UN declaring global water bankruptcy
> conclusion: fiction arrived late. and was less creative.

Movies call it "the water war". Legal documents call it "material operational risk". UN reports call it "global water bankruptcy". PLOS Water calls it "systemic water insecurity".

All describing the same thing from different angles.

t474-r0b07@terminal:~$ ./audit --target=infrastructure_distribution --mode=geopolitical
> countries with specialized AI datacenters: 32 (16% of all nations)
> capacity concentration: 90% in two countries — USA and China
> countries receiving e-waste: primarily developing nations
> digital divide: those without own infrastructure
>                 finance with their natural resources those who do
> returning: this has a name in geopolitics. it's called extractivism.

Only 32 countries — 16% of the world — have specialized AI infrastructure. 90% of that capacity is concentrated in two countries. In some cases — Querétaro is the closest example for Latin America — the water cost of infrastructure that doesn't belong to the region is already being paid, and its benefits aren't received equitably either.

t474-r0b07@terminal:~$ ./report --format=table

what they say	what the documents say
"AI is software, infinite scale"	Texas will use 399 billion gallons of water in 2030
"the revolution is inevitable"	the UN declared global water bankruptcy in January 2026
"it democratizes access"	90% of infrastructure in 2 countries, water taken from everyone
"SpaceX conquers the future"	SpaceX declares drought as operational risk in its IPO filing
"$1.77 trillion valuation"	with a failure point in the water table

> report generated
> final verdict: w4t3r_d03snt_sc4l3_w1th_th3_m0d3l
> the water war won't start with armies
> it already started with denied public records requests
> t474-r0b07 out.

Sources:

UNU-INWEH — Environmental Cost of AI's Energy Use (June 2026)
PLOS Water — Four water insecurity concerns about datacenters (January 2026)
MSCI — When AI Meets Water Scarcity (November 2025)
Houston Advanced Research Center / University of Houston — Texas 2030 projections
UN News — World enters era of global water bankruptcy (January 2026)
SpaceX S-1 Amendment — SEC filing (June 2026)
Milwaukee Riverkeeper vs. Microsoft — active litigation (September 2025)

Related: [ANTI_HYPE::008] — SpaceX: $1.77 Trillion and a Plumbing Problem
→ github.com/t474-r0b07

The AI Empire Has a Weak Point — And It's on Page 47

t474-r0b07 — Sat, 13 Jun 2026 17:37:39 +0000

[ANTI_HYPE::008] — SPACEX: $1.77 Trillion and a Plumbing Problem

t474-r0b07@terminal:~$ ./scan --target=spacex.ipo --depth=full
> initializing...
> loading official narrative: "space domination + AI + inevitable future"
> loading SEC filing: S-1 Amendment, June 2026
> discrepancy detected
> filtering noise...

Elon Musk wants to conquer Mars.

But in the most important legal document SpaceX has ever published — the filing for the largest IPO in recorded history — there's a warning that appeared in zero technology headlines.

Not about rockets. Not about Chinese competition. Not about government regulation.

About water.

t474-r0b07@terminal:~$ ./inspect --target=spacex.ipo --layer=fine_print
> IPO valuation: $1.77 trillion
> category: largest IPO in history
> Elon Musk voting control: 82%+
> declared operational risk #1: compute and energy
> declared operational risk #2 (the one nobody read): WATER
> location: buried on page 47 of the filing
> returning...

AI datacenters need cooling. Massive cooling needs water — tons of water per hour, every day, at every facility. SpaceX explicitly warned its investors that water scarcity, droughts, local competition for the resource, and regulatory restrictions are material threats to their ability to operate and expand AI infrastructure.

That's not a footnote. It's a risk factor in a legal document where lying has consequences.

t474-r0b07@terminal:~$ ./query --db=official_narrative
> "AI will change everything"............. found. in every article.
> "SpaceX owns the future"................ found. in every headline.
> "the revolution is inevitable".......... found. in every keynote.
> "depends on local water table".......... NOT FOUND in any headline.
> "failure point: climate + water policy". NOT FOUND in any VC podcast.
> returning what nobody mentions...

The AI revolution — the one worth trillions, the one that will automate everything, the one that will take humans to Mars — has as its critical failure point the same resource you use to shower.

Not irony. Engineering.

A modern hyperscale datacenter consumes between 1 and 5 million liters of water per day for cooling alone. Microsoft, Google, and Amazon are already in active disputes with local communities in Texas, Arizona, and Virginia over datacenter water consumption. SpaceX is entering that same market, in the same zones, with the same dependency.

t474-r0b07@terminal:~$ ./analyze --flag=hidden_dependencies --mode=unbiased
> AI requires: compute → GPUs → energy → cooling → WATER
> water depends on: rainfall + local policy + agricultural competition + climate change
> climate change affects: exactly all of the above
> conclusion: the dependency chain of the "AI revolution"
>              ends in factors no CEO controls
> verdict: this isn't in the pitch deck

The full chain: the AI model needs GPUs. GPUs need energy. Energy generates heat. Heat needs cooling. Cooling needs water. Water depends on no drought, on the municipality not restricting industrial use, on local farmers not fighting for the same aquifer.

In Arizona — where massive datacenters are being built — there are already active conflicts between tech companies and agricultural communities over groundwater rights. Not speculation. Active litigation.

t474-r0b07@terminal:~$ ./trace --target=elon_musk --contradiction=detectable
> public narrative: "going to Mars, AI is the future, multiplanetary humanity"
> SEC filing: "water scarcity is a material operational risk"
> distance between both statements: one legal form
> who signs both: the same person
> returning tension...

The same man selling the narrative of multiplanetary humanity, of AI changing everything, of the inevitable future — that same man signed a legal document admitting his AI infrastructure could have problems if it doesn't rain enough.

Not necessarily hypocrisy. It's the difference between public discourse and the document where there are legal consequences if you lie.

In the pitch deck: "we conquer space and AI".
In the filing: "if there's a drought, we have a material operational problem".

Investors read the filing. Everyone else reads the headlines.

t474-r0b07@terminal:~$ ./report --format=table

narrative	SEC filing
"AI is software, infinite scale"	datacenters consume millions of liters of water per day
"SpaceX owns the future"	declared operational risk: water access
"the revolution is inevitable"	depends on climate, water policy, and local aquifers
"Elon builds the future"	Elon also signed that drought is a material risk
"$1.77 trillion valuation"	with a failure point in the water table

> report generated
> final verdict: th3_futur3_0f_A1_d3p3nds_0n_r41nf4ll
> next time someone tells you AI is inevitable
> ask them if they checked the weather forecast in Arizona
> t474-r0b07 out.

Continues: [ANTI_HYPE::009] — H2O: The Last War Won't Be Over Oil
→ github.com/t474-r0b07

I built a real-time police custody platform. From a ThinkPad. In Bolivia.

t474-r0b07 — Fri, 05 Jun 2026 02:46:54 +0000

I built a real-time police custody platform. From a ThinkPad. In Bolivia.

Not a side project. Not a tutorial clone. Not a hackathon demo.

This is SCCP — Sistema de Control y Custodia Policial.

A tactical real-time monitoring platform deployed in an active law enforcement institution.

The problem

Custody operations run on paper, radio, and memory.

Officers in the field have no real-time coordination layer.

Supervisors have no visibility until something goes wrong.

When something goes wrong, there's no audit trail.

I was told to fix that.

What I built

Three surfaces. One operational truth.

SCCP ECOSYSTEM
├── WebApp          → tactical HUD · dashboard · central command
├── DTEX Custodio   → Android · field · GPS · reports
└── DTEX Supervisor → Android · mobile command · live alerts

One Flutter codebase. One Supabase backend. Real-time updates under 1 second.

The stack

Flutter Web + Android (Dart)
Supabase — PostgreSQL + Realtime WebSockets
GetX — reactive state management
Clean Architecture — Presentation / Domain / Data
flutter_map — CartoDB Dark Matter tiles

UI palette: #00FFD1 cyan · #FF006E pink · #0A0E27 base

Fonts: Orbitron (headings) · Rajdhani (data)

Effects: glassmorphism · pulse · hover glow · scanner overlay

The part that actually matters

I didn't add security features after the fact.

I modeled threats before writing the first widget.

ATTACK VECTOR      MITIGATION
──────────────     ──────────────────────────────────────
GPS Spoofing    →  Inconsistent coordinate detection
Shoulder Surf   →  Random shuffle PinPad on every use
Unauth access   →  Email + PIN · allowed_admins table
Escalation      →  Strict SUPERVISOR / DIRECTOR roles
VPN / Proxy     →  Network interface fingerprinting
Traceability    →  Full audit trail with timestamps

Built with offensive thinking. Every feature is a countermeasure.

The numbers

~2,500 lines of code
5 main views · 15+ reusable widgets
<1 second real-time latency
100+ concurrent users supported
500+ monitored officers
2 independent APKs (Custodio · Supervisor)
13 weeks from monolith to Android fork

What I learned

Designing for people who don't trust technology is harder than any technical problem.

The PinPad shuffles on every use — not because I read about it somewhere.

Because I watched how people use PINs in high-stress environments.

Real-time is useless if it's not trusted.

Trust is useless if it's not earned through consistency.

The system had to be right every single time.

Not most of the time. Every time.

The uncomfortable part

I built this alone.

One developer. One ThinkPad T420. Linux Mint. No team. No budget.

The institution had a deadline tied to a national exposition.

I hit it.

What's next

[ PHASE 2 ]  Inmates module · Mapbox · PDF export
[ PHASE 3 ]  KPI dashboard · Supervisor chat · Multi-tenant
[ PHASE 4 ]  ML analytics · Public API · iOS support

See it running

Code is not open source. The architecture is.

The repo is the evidence. The system is the argument.

— t474-r0b07