19 most EXPENSIVE Tech Fails of all time

Tech for Change
4 followers
1 Video View·May 8, 2026
We’ve never put more effort into a tech fails video than this one, please tell me it was worth it 🙈 Go to https://surfshark.com/boss or use code BOSS at checkout to get 4 extra months of Surfshark VPN!
I'm a French jewel thief, but I got a bit of a problem.
I don't have any jewels.
So I'm off for a little day trip to the Louvre to get some.
The thing is though, the Louvre is world's most visited museum.
It's home to humanity's rarest treasures like the Mona Lisa
and even the French crown jewels.
And so to actually reach this security would require
an absolute black ops, cover of night,
oceans 11 level operation.
Well, maybe not.
Because in October, 2025, four thieves literally committed
daylight robbery at the museum.
They rocked up to a window in a furniture lift
wearing high-vis vests during opening hours.
And they just took the crown jewels.
They hopped back into their lift,
descended comically slowly back down to the ground
before disappearing away on their scooters at 9.30 AM.
We're just in time for a breakfast croissant.
See, the Louvre had a bit of a tech problem.
Their entire cybersecurity system was based on software
almost as old as me.
With a 2014 audit finding that they were still using
Windows 2000, which was well past the point
of still getting security updates.
And yeah, they did have CCTV.
It was just facing the other direction.
So they had no view of this balcony being used
to access the window.
Okay, but still these thieves would have had
to be something special, right?
To be able to identify this critical blind spot.
Oh, that's right.
The password to the entire video surveillance network
for the Louvre was Louvre.
It was not a joke earlier, by the way.
This actually happened.
Allowing them to drive off into the rush hour traffic
with $102 million of jewels.
The Louvre's president, Laurence Descartes,
was so appalled that she tried to resign the very next day.
And I would still only call this a five out of 10 tech fail.
Because it got me thinking,
if an easy-to-guess password and four dudes
in a furniture lift are enough to cause $102 million
worth of damage, then how expensive can it get
when the mistakes get bigger?
That's what I found out.
And I kind of wish I hadn't.
Like, you know Dyson, the company who makes your vacuum,
your hand dryer, your hairdryer?
Essentially, if it moves air, then Dyson will sell it
to you for like $400.
Well, in 2016, James Dyson decided that
what he really wanted to move was people.
So he started secretly building an electric car.
And look, it wasn't a crazy idea.
Dyson specializes in high-performance batteries
and electric motors.
And that's exactly what an EV needs,
which is why they went big.
Developing a battery pack that could go 600 miles
on a single charge.
We're still not seeing 600 miles now.
And this was 2019.
Not to mention a floating hologram heads-up display,
seven seats, and a design that I think
I'm physically attracted to.
Oh yeah, and the seats were ergonomically redesigned
from scratch because Mr. Dyson personally hates
the lack of lumbar support in normal cars.
You can kind of see his point, to be fair.
Not our most comfortable getaway.
The slight problem is Dyson ended up reworking so much
that they eventually realized just to break even,
they would have to sell each of these cars
at today's equivalent of $275,000.
And it took our good friend James burning through
$900 million of his own personal money
to come to this realization.
At which point he had no choice
but to scrap the entire project,
go home and sit on a chair
that probably gave him back pain.
Six out of ten.
Lads, we want some lunch.
But sometimes the biggest cost of a mistake
is to your reputation, like it was with Taco Bell.
Because in 2025, this company implemented
an AI-powered drive-thru system
in over 500 of their restaurants,
hoping to improve customer experience.
And, um, let's just say it wasn't very good
at taking orders.
Oh, and my absolute favorite is when this guy asks,
can I get 18,000 water cups, please?
What can I get for you?
The way that the system just instantly dies
and hands it over to a human worker is class.
And just a repost of this AI breakdown on YouTube
is currently sitting at nearly 30 million views.
And so, responding to the fact
that their AI's greatest achievement
was becoming a meme,
Taco Bell's chief digital officer, Dane Matthews,
told the Wall Street Journal
that they were going to think carefully
about where not to use AI going forward.
Gee, thanks, Taco Bell.
How did that not occur to you before?
Having just watched McDonald's
cancel their own AI drive-thru
after it went viral for putting bacon on ice cream
and nine sweet teas in one order.
They keep trying, they keep failing,
and in the process,
they just keep funding their own funerals.
Where's the rest?
That's all they had left.
What about yours?
Oh, I had a croissant.
Right.
We should probably go then.
But it's more embarrassing for Will Smith,
who used AI to replace people,
but just didn't even realise.
Last year, he posted a video from his music tour,
and do you notice anything a little off about it?
Specifically, the crowd
that has clearly been edited with AI.
Kind of looks like Will wasn't quite satisfied
with the turnout,
so decided to embellish it a little
with some AI-generated concert goers,
hoping no one would notice.
Big fan of the guy, so moved by Will's music
that he's wiping his tears through his glasses.
The sign saying,
from West Philly to West Swiggy.
And especially the one claiming
that Will helped them survive cancer.
The idea of all of this being AI-generated
was horrific news for Will Smith's,
let's be honest,
already waning reputation.
With every comment,
some sort of joke at his expense.
Like, unreal concert man.
Literally.
And Will Smith has not only
just melted his fans' hearts with his concert,
but also melted their entire bodies.
But the worst part,
the entire fiasco was an accident.
Turns out, the crowds were real.
You can see them in phone footage from the gigs,
and bafflingly,
in photos posted by Smith himself,
where you can directly cross-reference things
like the Swiggy sign was there
in his Switzerland show,
although it actually says Swizzy.
So most likely,
Will's social media team
just used an AI tool
to put together the highlight reel.
They fed in genuine photos and videos,
but because the tool itself was AI,
it introduced all of these
unintended side effects.
Either way, damage is done.
Everyone thinks Will Smith
faked a crowd to stoke his ego.
He's the AI crowd guy now.
Four out of ten.
Oh, that's hot.
That's hot.
But speaking of men with image problems,
you've probably seen Elon Musk's
Tesla Optimus humanoid robot at this point,
even if it was only as a guy in a costume.
Musk has stated that the Optimus robot
will eventually account for 80% of Tesla's value,
which, I've got to say,
sounds like a ludicrous prediction.
He's also claimed that Tesla currently
has two of these Optimus robots actually working.
So when you see one out here
serving drinks at a product demo,
you're kind of expecting the real deal, right?
Well...
Um...
The fall was embarrassing enough,
but it's actually the hand movement
that got people talking.
Look closer.
Doesn't it look an awful lot
like it's removing a headset?
Kind of like the thing
was actually being remote controlled
by, say, a human operator.
And Tesla have a track record
of doing exactly this.
Putting human-operated robots out there
and deliberately being very lax
about letting people know
that these are not fully autonomous.
Add in the fact that Musk is currently
chasing a $1 trillion payout from Tesla,
where one of the goals that he needs to hit
is to ship a million bots,
and it starts to make sense
why he's doing way too much
to convince you that the future is now.
Maybe just hang on
until you have a working prototype
before showing it off.
At least that's what Nothing did
when they wanted to impress customers
with the camera on their new flagship phone 3,
holding a bunch of demo events in stores.
Five photographs were shown off
on these in-store demo units
with the text,
here's what our community has captured
with the phone 3.
Do you want to know what's crazy?
Zoom into the reflection of this one
and you can actually see the DSLR camera,
not phone 3, that took it.
And then with a little internet sleuthing,
fans came to realize that in fact,
every single one of these shots
was in fact a publicly available stock photo.
Nothing eventually came out to claim
that they were just placeholder images
that they had intended to replace,
but that the units had gotten to stores
before that happened.
Malicious or not though,
this is a pretty costly,
extremely avoidable PR moment
to be having while being the underdog
just as you release your first ever
flagship smartphone.
But Nvidia has definitely lost a lot more rep
and didn't have a huge amount to lose
in the first place,
ever since they pivoted their focus
towards providing graphics cards
to giant AI companies,
kind of ignoring their original customers,
the gamers.
But then instead of deciding
that it was time to listen to the players,
Nvidia just decided to go full-on friendly fire
with DLSS 5.
So DLSS,
which stands for Deep Learning Super Sampling,
has been one of Nvidia's superpowers
for a long time.
It's basically a smart graphics technology,
which means that instead of each generation
having to keep doubling the amount of hardware
they're giving you,
they can instead upgrade the resolution
and the frame rates in your games
using clever machine learning tricks.
However, the latest version, DLSS 5,
goes one step further
to upgrade even the lighting
and the textures too,
which sounds fantastic, right?
Who doesn't want a prettier game?
But let's actually think about it for a second.
This scene of Grace
from Resident Evil Requiem
is a scene of her heading
to investigate the house
where her mum was murdered
in front of her eyes.
But Nvidia's AI slop filter
doesn't know that.
It just sees a face
that looks a little dark and traumatized,
uses its training data,
which tells it that faces look better
when they're yassified with blush and eyeliner
like your resident ego,
but in the process,
completely flies in the face
of the possibly months
that the developers spent
hand crafting the very deliberate way
that they wanted her to look in the scene.
Or why does the professor
from Hogwarts Legacy
need more wrinkles?
Guys, it really feels like
this is just AI seeing an older person,
then going through
the thousands of stock photos
it's been trained on
to make her more old.
It's funny because CEO Jensen Huang
responded to the overwhelming backlash,
arguing that it's not about
just putting an AI filter over everything
and that instead,
DLSS5 is meant to be
integrated with the artist.
And so it's about giving the artist
the tool of AI.
Which made it extra awkward
when it came out
that the developers of these games
found out at the same time
as the public
that their games were being
altered in this way.
5 out of 10.
Nvidia's gamer cred
was already in the bin.
Now it's in hell.
And while we're down there,
there's a few fails so far
where there's been some mystery
as to whether or not
it's been a deliberately
perpetrated crime or not.
Not so much the case with ransomware,
a type of malware that locks you
out of accessing your data
so that criminals behind the attack
can sell it back to you.
But at least there are companies out there
like Digital Mint
who specialize in negotiating
those ransom payments down.
Right?
You know, people who truly understand
the criminal mind.
Yeah, about that.
In 2023, it was actually employees
from inside these companies
who used this specialized knowledge
to carry out their own attacks.
They targeted at least
five American firms.
They stole the data
and demanded millions in return.
But this is where
it gets truly unhinged.
When those victims panicked
and they called Digital Mint up for help,
who did Digital Mint
assign to the cases?
Well, none other than
one of the guys that attacked them.
Angelo Martino,
who in basically
the human embodiment
of the evil Kermit meme
had managed to put himself
in a position where he was
playing both sides
and negotiating with himself.
This must have been
some tough negotiations
because he managed to get
all five of these companies to pay up.
One of them even paid
26.8 million dollars
and it was a non-profit.
What?
Definitely a non-profit now.
Can you believe that just months before
Digital Mint featured
one of these guys
in their employee spotlight?
And I quote,
we bridge the gap
between good, hardworking people
and bad actors.
They had no idea
how right they were.
And basically,
everyone lost here.
The ransomed firms ended up
paying out 75 million dollars,
no one's ever going to
call up Digital Mint again
without thinking twice,
and the three employees involved
have been charged
with up to 20 years in prison each.
6 out of 10 fail.
But where it goes
beyond funny
into just straight up terrifying
is when a man named
Sammy Azdufal
developed an app
to allow him to control
his DJI Romo robot vacuum
using a PS5 controller,
simply because he said
sounded fun.
But that's not the scary part.
Unfortunately,
Azdufal didn't realize
his own coding power.
The custom remote control app
he built pretty quickly
using clawed AI code
accidentally granted him control
of over 7,000 DJI robot vacuum cleaners
across 24 countries
the second it connected
to DJI servers.
How?
Because instead of DJI
encrypting the data
from its devices
like they should have,
it was literally laid out to him
in plain text.
Zero authentication required,
allowing him complete access.
So, okay,
he could now vacuum
anyone's house on a whim.
That feels like a
Doofenshmirtz level scheme.
Evil, but ultimately harmless.
Until you realize that
these robots all had
cameras and microphones,
and that he also happened
to have access
to the live feeds from them.
Not to mention the floor plan
of each house
that the vacuums had mapped out
with all of their sensors
and their location too
via their IP addresses.
Thankfully, he was a nice guy about it.
He immediately reported
the floor to DJI
and they immediately rewarded him
with a 30k payout.
But just imagine
how catastrophic
this could have been
if he wasn't a nice guy.
Remember, as Doofal
didn't hack into DJI servers
or do anything complex,
he simply became
god of all robovacs
by accident.
And it's just mad to think
that someone can so easily
stumble into so much
personal data from you.
At least when you're using
Surfshark VPN, our sponsor,
anyone who might be snooping
isn't getting anything useful
because your identity is masked.
And they just made it even crazier
by launching this
brand new custom standard
called DOSOS.
Instead of cramming
everyone's data
into one shared tunnel
like a traditional VPN,
DOSOS gives you
your own dedicated
private data lane.
And the reason that I
bang on about Surfshark
is because it's like
the one deal in tech
that feels like a steal.
With the code boss,
you can literally get Surfshark
for $1.68 a month
because they're celebrating
their birthday.
And that would cover you
plus an entire crowd
at a Will Smith concert,
if you wanted.
Now at least Sammy
had the decency to feel bad
about his accidental data theft.
Can't say the same for chat GPT,
which is meant to
revolutionize search
by accessing everything
on the internet
and bringing it to you.
Now that would require
having the rights
to a lot of content
that would end up expensive
and time consuming to acquire.
So naturally OpenAI
just ignores that part.
Why?
Go fast and break things,
of course.
Who needs the law
when you have a ridiculously
ballooned valuation?
So it's a shock
to absolutely no one then
that a group of authors
and publishers
are suing OpenAI
for copyright infringement.
But it's what's happened
as part of their investigation
that's been the dramatic upset
because the group managed
to acquire leaked Slack messages
and emails from OpenAI
in which their employees
openly discussed
the mass deletion
of two data sets
the AI was trained on
that they knew
consisted of pirated books,
hilariously named
Books 1 and Books 2,
by the way,
in case it was unclear.
So not only did they have
other people's pirated data,
but they knew full well
that what they were doing was wrong
and tried to dispose of the evidence.
So a New York district court
has now ordered OpenAI
to hand over those messages.
And if those messages
demonstrate willful infringement,
and I don't really see
how they couldn't,
this could take the damages
anywhere from $750
per piece of stolen work
to possibly $150,000 per work.
An insane amount
on the scale of data
that these guys are working with.
We don't know the total fines
they could face yet,
but we can get some idea
from a recent lawsuit
against Anthropic
for similar copyright infringement,
which saw them settling
for $1.5 billion.
Settling.
7 out of 10.
But while the exact consequences
of that are still up in the air,
OpenAI has recently had
an even bigger oopsie
that's already cost them
some very real, very large dollars.
See, in 2025,
the company announced
a noble new mission
that doom scrolling
wasn't bad enough as it was,
and that what humanity really needed
was an entirely new
short form video app called Sora
that lets you doom scroll content
that was entirely AI generated.
A slop talk, if you will.
And to kick off the fun,
while downplaying
the dystopian impending threat
of anyone being able
to create a deep fake
in like two clicks,
Sam Altman gave all users
global permission
to create videos
using his own likeness,
which of course immediately backfired
with a litany of embarrassing videos
mocking the guy.
Like this one of him
begging for GPUs
at a doorbell camera.
I can't train anything.
Please, if you have anything,
A100s, 3090s, I'll take them.
Physically stealing art
from Studio Ghibli.
Get them back!
Nope, too late.
Hey, come back here!
Free art, baby!
Or this one of Altman
hosting a Hunger Games
style competition,
forcing contestants to fight
over literal slop.
And then of course this,
which I present without comment.
Meow.
Turns out Super Sam
did this to himself
for absolutely nothing
because Sora shut down
just six months after starting.
Let me try and put into perspective
just how much of a flop
this slop was.
Sora generated in total
Sora was costing them
at peak usage
around $15 million
per day to run
all while the user base
was collapsing under their feet
with a 66% drop
in just the first 90 days.
And as if that wasn't
already enough damage,
OpenAI also,
in the process of losing Sora,
lost an investment deal with Disney.
Disney was going to pay OpenAI
for a stake in the company,
which would have also given
Sora users access to
200 plus characters
from across the entire Disney universe.
How much was Disney
going to pay them?
$1 billion.
Now you already know
that in 2022,
Elon Musk purchased Twitter
for mountains of money.
We've been living
with the consequences ever since.
But there is one consequence
that you might've missed.
That Musk is now being sued
by the investors who originally
put their money behind that deal.
And the reason is
one of the funniest things
I've ever seen in a courtroom.
So do you remember when Musk
was trying to wiggle out
of buying Twitter?
He'd already committed at this point,
but he was trying to pull away
because he said
too many of its users
were actually bots,
not people.
And by the way,
he said this on Twitter,
publicly,
in tweets.
Well, those tweets
tanked Twitter's stock price.
That made all of these investors panic.
They believed him.
They assumed that the whole thing
was falling apart.
And so they hurriedly
sold the shares they had,
losing millions
compared to what they paid for them.
But the deal wasn't off.
Twitter took Elon to court
and forced him to buy Twitter anyway
at the original agreed price
of $54.20 per share.
But all those investors
who sold their shares at a loss,
they never got their money back.
Which leads us to now,
where a California jury
has declared Musk liable
for misleading them
with his own tweets.
Tweets which he himself described
under oath
as stupid tweets.
You just can't make this stuff up.
The man who bought Twitter
to protect free speech
is being sued for what he freely said,
stupidly,
for up to $2.6 billion.
And if you think that's a lot of money,
then you're not ready
for the metaverse.
Do you remember when Facebook
rebranded to meta
out of nowhere
and made a massive bet
on our near future
becoming almost exclusively
virtual reality?
The hub for which
was meant to be Horizon Worlds,
an online VR world
where you can hang out
with all your other friends
who definitely have a Quest headset.
Now, because they were
trying to take VR
from merely something
that people use to play games
into this alternate reality
where we'll all one day
live and work
as well as play,
it was important
to have the infrastructure
for users to visit different worlds
and create their own
for whatever use they might want.
Like for instance,
taking a fake selfie
in front of a fake 240p Eiffel Tower
while trying not to let
the fake existential dread
creep in too much.
Wait, no.
Dread was real.
Well, it seems like
meta are starting to catch up
with the rest of us
who already know that
no one's interested
in hanging out in VR,
whether it gives us legs or not.
Because this year,
they announced that
they were removing Horizon Worlds
from the Quest app store,
deciding instead to focus
100% of their effort
on mobile instead.
Because yes,
this is exactly what
the phone user base
has been yearning for.
But then a glimmer of hope
for the roughly
four remaining Horizon Worlds fans,
meta did a U-turn
and announced that in fact,
no, we're going to be
keeping VR access
to the platform after all,
just according to them
to support the fans
who've reached out.
Must have been a long day
going through all of those emails.
But Horizon Worlds,
as embarrassing as it is,
is just one tiny part
of the metaverse's failure.
Because amidst the fallout,
we've also come to learn
the total amount of money
that the whole metaverse project
has lost meta.
And you're going to want
to sit down for this one.
We've seen millions this video.
We've even seen a couple of billion.
But meta burned through
80 billion dollars.
That's pretty much what it would cost
to end world hunger for a year.
What did it get spent on?
This.
Cheers, Zuck.
I've got to get out of here.
