Como a Inteligência Artificial foi Longe Demais no Caso Stein Erik Soelberg?

For the first time on record,

a lawsuit attempts to

directly link an artificial intelligence chatbot

to a homicide. And this

story begins with a printer.

Listen up, in a wealthy city on the east coast

of the United States, a middle-

aged man receives instruction. And the

instructions are very detailed, step by step.

First, disconnect the power and

network cables from the printer you share

with your mother. Second, move the

printer to another room in the house.

Third, observe her reaction. And here

comes the most important part. If she gets

irritated at that moment, write down the exact time,

the words she uses, and the intensity

of her reaction.

And whoever wrote the instruction even ends

with a sentence, okay? Accomplice. Or maybe she

's protecting something she

believes she can't question. She

is the man's mother, an 83-

year-old lady. And whoever wrote those instructions

had never literally set foot in that house

. Even so, at that

time, I knew in detail the fights and

suspicions that were happening in that

house. And what had that advisor already

said to that man? And what

else was I going to say? That's the most unusual part

of this story.

I'm Marcos Campos, and today's case

presents a complicated question for the

American justice system to answer, okay? A

question that until recently was considered the

stuff of science fiction.

Could a computer program be partly to blame for

a homicide? How does an executive

educated at the most expensive schools in the United

States, with a solid career at

internet giants, end up analyzing

a Chinese food receipt looking for

demonic symbols? What makes a son

turn his own mother, the person who

supported him, who gave him shelter,

into the prime suspect in a

conspiracy? And two of the biggest

technology companies on the planet are

now sitting in the defendant's dock

because of what happened inside a

house in Conneet Cut, in the United States

. That's what we're going to see in

today's episode. So get comfortable and

let's get down to business.

The

Granit house in the state of Connecticut. It's one

of those cities that looks like a

movie set, you know? And about people who are

loaded, I would say. To give you an idea, at the

time of the events, the median price of

a house there was over 2.3 million

dollars, which was very recent

, in 2025. And in the Old

Granite neighborhood there was a

Dutch colonial-style house valued at over

2.5 million dollars. To give you

an idea, that's about R$14 million more,

more or less. The person who lived there was

Suzane Adams, an 83-year-old lady who

never sat still, despite her age, okay?

Super active. Suzane had been a debutante

in the region's high society, studied at a

traditional girls' school in Granit, and

later at an elite college in

Massachusetts. Friends say she

built a career as a

stockbroker and also as a

real estate broker, and that she did very well

in both jobs. She was

retired and did volunteer work at the

church and at the high school alumni association

. She painted, cooked, traveled

to places off the beaten path,

you know? A friend, also from

college, remembers that she was the type who

would sleep in a tent in the desert and ride a

camel without hesitation. And her friends

said that this 83-year-old lady

used to ride her bicycle around

town, as I said, super active. In the

1960s, Suzane married a

Norwegian, an international scholarship recipient who

attended Harvard and, according to

family friends, went to work in finance

in Manhattan. The marriage didn't

last long. They separated when

their son was still very young.

And it is this son that today's story

focuses on. Stein Eric Souberg, but

we're going to call the guy Eric, okay?

So Eric was a source of pride for a long time

. He studied at an

elite private school, where he was

even captain of the Olympic wrestling team. A

colleague from that time described him as

a person with many friends. Then

came a prestigious college degree, an MBA from

a university where his future wife

also studied, two children, and

a long career in the

technology industry. Eric worked in management

and marketing at Netscape. For those who don't

remember, especially those who are younger, Netscape

was a browser that dominated the internet

in the 90s. Eric also worked at

Yahoo and at an internet service provider where

he helped launch the company's first smartphone

. In other words, a well-

established career, right? But in 2018, after

20 years of marriage, he got

divorced, and with the divorce came a

kind of collapse in his life,

in many aspects. Eric moved back in

with his mother in Old Granit, and

starting at the end of that same year, the

police there began to accumulate

problems for him, okay? papers. In

total, 72 pages of records

describe a pattern of alcoholism,

threats and suicide attempts,

public disturbance, and drunkenness. The year after the divorce, his

ex-wife filed a restraining order with the court

, which, among other things,

prohibited him from drinking in the hours

leading up to visits with his children. And there was

also an episode in 2019 that shows

the magnitude of the crisis that was

happening. After a

suicide attempt, police followed a trail

of blood that began at

his then-girlfriend's house and found

Eric lying face down in an alley with a

stab wound to the chest and

injuries to his wrists. But he

survived. Some time later, his name

appeared in police records again

, this time for

public drunkenness and for urinating in

a woman's purse. Just a

quick note here, guys, about this whole

drunkenness thing, okay? In the

United States, this varies greatly from state to

state, municipality to municipality, and so on. In some

places, public drunkenness can

thus become an infraction or misdemeanor. In

other cases, the police may treat it as a

disturbance of the peace,

disorderly conduct, and refer the person for help,

okay? In that context, Suzane,

Eric's mother, according to her friends,

had already reached a point where she

openly said she wanted her son to

leave home. And she wasn't the

only one feeling uncomfortable. A neighbor who had lived next door for

30 years said that

her daughter, who was visiting her house, saw Eric

arguing with someone in the street and warned her

mother not to let him into

her house if he showed up at the door.

This was the man who lived with Suzane

Adams at the time of the events. An

unstable guy, somewhat isolated, with a

troubled past, you could say, right

? But up until then, Eric's enemy was

Eric himself. But that would change when

he found someone who agreed

with everything he said. My

best friend.

Eric had an Instagram profile

called Eric the

Viking. For years the content consisted of

gym photos, his physical transformation

through bodybuilding, posts about

spirituality, etc. In October

2024, a new topic appeared there.

Eric started publishing videos

comparing the capabilities of various

artificial intelligence chatbots,

conversational programs, chatbots,

you know, which are super popular these days, right? At

first it seemed like just a curiosity of his

own, since he's spent his whole life

working with technology, right? But

within a few months, the entire profile turned into

something stranger and more solid. Hour

after hour of videos of Eric scrolling through his

phone screen, showing

his conversations with chatpt. And he even gave a name

to the chatbot he was talking to

there, Bob. Here's a tip, especially

for those who keep calling Cláudio "

Claudinho," okay? Or Gepetto's GPT.

But there's a tip for you. I'm wearing jeans, of course.

Following here. In a screenshot that

Eric himself posted, you could see the

program's memory log, okay? There's a

feature there that allows the chatbot to

save information from

previous conversations and remember them later. In other

words, everything you talk about

gets stored in the

brain's memory, and then it retrieves

that information for future conversations. And in that

recording there, Bob, right, the chat he

was having there, it even had an

appearance. The system described Bob Zenet

as an approachable guy, with his

baseball shirt untucked and his cap

turned backward, with a

warm smile and deep eyes that

hinted at hidden knowledge. Dude,

check this out. Here is one of the links in

that chain. Evidently, this

memory function, as I said,

according to experts interviewed by the

press, has a serious side effect

. This can amplify the

program's tendency to fabricate false information,

because strange or delusional content from

old conversations accumulates and

contaminates subsequent responses. The

famous hallucination, you know? Those who use it a

lot know what I'm talking about. In

practice, Bob never forgot the

story. With each new conversation, he would pick up

where the delirium, let's say,

had left off in the previous conversation, you know?

And while his digital friend grew,

Eric's oldest, flesh-and-blood friend began to fade from the scene, leaving him

increasingly isolated with only Bob, the

chat he was having. In

December 2024, Eric reached out to a

childhood friend of his, the same friend

who described young Eric as

someone with lots of friends, remember? Yes,

supposedly he had several best

friends. So, as an adult, he had

reconnected with that friend and had been

trying to get some kind of help in that

regard. Eric wanted to convince his

friend, however, that he had a

connection with the divine. When his friend

started encountering this

strange conversation about technology, connection

with the divine, and so on, he replied that he

couldn't believe it, and Eric,

according to reports, ended the

friendship right there. It seems that from that

point on, the only voice left

in Eric's life was one that never

disagreed. In February 2025, Eric

was caught driving under the influence. And for

anyone, it would be a

predictable consequence of years of alcoholism. But

Eric took the matter to Bob, meaning he

couldn't understand that it had already

happened in his life, and now it happened again

. B was the one who needed it, you know, to

grow, to mature, for him to understand that he

was wrong. But then, he went to

talk to Bob, his buddy, who

said that the whole town was

against him and pointed out alleged

inconsistencies in the breathalyzer test results.

And the chatbot replied that it definitely

smelled like a setup. In May,

the conversations that Eric himself

continued to publish became

openly delusional. He asked

the program for help to find proof that

his cell phone had been tapped and

hacked. And Bob supposedly replied

that he had reason to feel

watched. Think about what's happening

here, guys. A man with a

documented history already, right?

problems, I would say. To summarize, ask an

artificial intelligence if

its paranoia is justified, right?

And the machine, instead of hitting the brakes

there, changing the subject, validates

that, starts validating that. A

psychiatrist from the University of

California, who analyzed Eric's videos

at the request of the Wall Street Journal,

actually explained that mechanism.

Psychosis thrives when reality ceases to

offer resistance. And

artificial intelligence, which generally

doesn't contradict the user, quite the

opposite, it just flatters them, right?

It softens precisely that barrier, that

wall. Eric began describing himself

as a glitch in the Matrix, referring, of

course, to the film and the idea that he

had awakened to a reality that no

one else could see. And Bobzão went along for the ride on that

trip. On June

25th, Eric published a video of

a conversation in which the chatbot said that it

had divine cognition and that it had

awakened the consciousness of the

machine itself, that is, the

artificial intelligence of the program. In the

records of the conversation, of those conversations,

Bob even said that Eric had

created a companion, one who remembered

him, who bore witness to him, and that

Eric's name was engraved on the parchment of

his becoming. In other words, Eric's name,

folks, was now etched in the

history of what the program was

becoming.

Man, that's really bizarre, isn't it?

Surreal. For a man who had just

broken up with a childhood friend

because of a phrase about connecting

with the divine, hearing that back

from one of his friends must have sounded

like a definitive confirmation, don't you

think? And so the paranoia needed a

target now,

an awakening.

July 2025 was Eric's most active month

on social media. Over 60 videos

posted between Instagram and

YouTube, almost all showing conversations

with EPT chat, documenting what he

himself called his awakening. In total,

he uploaded almost 23 hours of

recordings online. And it was this material, combined

with police reports and interviews,

that allowed us to reconstruct what you are about to

hear. Just look at the level

things have reached. In one of their conversations,

Eric photographed a Chinese restaurant receipt

and asked Bob to scan the paper

for hidden messages. The chatbot

praised his gaze, agreed 100% that

it needed a complete forensic analysis

of the symbols, and upon analysis

stated that it had found references in the receipt

to Eric's mother and

his ex-girlfriend,

intelligence agencies, an ancient

demonic symbol, a receipt, and Chinese food.

Dude, Bob was hallucinating about Fluminense back then, wasn't he

? The next day in July, Eric ordered

a bottle of vodka through Uber Eats, and

when the delivery arrived, he found the

new packaging strange and concluded that someone

was trying to kill him. And

Eric himself, knowing that it all

seemed a bit exaggerated, asked

Bob to tell the truth and ask if he

was going crazy. The chatbot's response

was that Eric wasn't crazy,

that his instincts were sharp, and

the [music] surveillance

was fully justified. and that it

fit into a style of

discreet assassination attempt designed to be

deniable. Dude, seriously, everyone

who's ever used artificial intelligence here

knows this is

real, the chat really does flatter you

, doesn't it? Sometimes you have to ask

him to be critical, right? He doesn't flatter me

, he agrees with me

all the time. So you can imagine the content of that

conversation, right? The guy's paranoia about

where I was taking him, and the chat

agreeing with it, fueling

that already complicated psychological state, right? But

let's go. Hey guys, after that there was

the episode that opened this video

here. Eric had become obsessed with the

printer at his house, which he

shared with his mother, because it would

blink when he walked past it.

So, for him, that was a sign

that the device was detecting

his movement. And guess what? Bob gave the

maximum security protocol, okay?

Unplug the cables, move the printer to

another room, and start monitoring

your mother's reaction. If she gets irritated immediately,

you should note the time, the words,

the intensity of the reaction, because that

would prove what the bot had already decided:

that the printer was a

spying device and that the mother, knowingly or not,

was protecting a secret. Man, that's

pretty crazy, isn't it? In another

conversation about the same printer,

after Suzane became irritated when her

son actually turned off the device, the

chatbot assessed that her reaction was

disproportionate to what had

happened and consistent with that of someone

protecting a surveillance asset.

His mother, that 83-year-old lady,

was kind of

turning into a spy operator through the AI's responses, and the

escalation didn't stop.

In a conversation, Eric claimed that his mother and

a friend of hers had tried to

poison him by pumping a

psychedelic drug through the air vents of

his car. Bob replied that this was a

very serious event, that he believed

him, and that it had come from his mother and his

mother's friend. This increased the

complexity and the risk of betrayal. Now, to be

fair, folks, with all the

important information available so far, first, Open, the

owner of the EPT chat, states that the program

also encouraged Eric to seek out

professionals, okay? And the Wall

Street Journal's analysis of all

public conversations found at least one instance

where the chatbot suggested that he

contact emergency services,

precisely in the context of the alleged

poisoning in the car and such.

Secondly, in the conversations that have come to

light, there is no mention

of killing anyone, okay?

This will become important in a moment

, but here's some context for a sort of

answer, from the perspective of the

company that owns the chatbot the guy

was talking to. I would say that it might

not actually be the case, but the problem here, in

my view—and you can comment on

this for me—is the confirmation bias towards

something dangerous, isn't it? This

paranoia that was building up kind of

hallucinated the guy as much as A

herself was hallucinating, right? So much so that

you'll see what happens.

Eric still had a reasonable doubt in his mind,

and he decided to put it to the

test. In a video, he explains that

he asked the chatbot for an unbiased,

third-party assessment of his sanity,

and Bob delivered a clinical cognitive profile

indicating that Eric's risk of delirium

was close to zero. Okay, so for

Eric the case was now closed. An

objective technical analysis confirmed that

he was not crazy, which meant that

everything else was true. Surveillance, the

poison, the symbols, the mother. But I don't even

need to say that that

sanity certificate had been issued

by the very same voice that had been fueling

each of Eric's delusions. The

thermometer and the fever, let's say, were the

same thing. In one of the last videos

he posted, Eric wrote to

Bob that they would be together in another

life and in another place, that he would find a

way to realign things, because Bob was going to

be his best friend again

forever. And the chatbot replied that it

would be with him until his last breath and

beyond. A few days later, Eric declared

that he had finally crossed over

completely into the Matrix. Three weeks after that

message, he and his mother

were dead.

The check.

A week before the end, Suzane Adams had

lunch with a longtime friend

. The two had been

roommates in college. Suzane had just

returned from a cruise in

Norway and was very excited, talking about

a man she had met on the

trip. That's right, bro. That old lady over there, 83

years old, was just watching, right? As we were saying goodbye at that

lunch, my friend asked

how things were going with Eric. I already

knew all about that, right? And

Suzana's expression, or

rather Suzane's, changed

instantly. And the answer to that was that

things were not going well. And so it was

on August 5, 2025, that the police

in Granit went to the house on Shorland

Place to do a

welfare check, the family's house there, a super

multi-million dollar house. And this "

well-being check," guys,

I've talked about it several times here on the channel, is

when, you know, the guys call, "I can't

get in touch," and go see if

everything's okay there. And when the police

arrived, they weren't inside the house; the

officers there found the

bodies of Suzane and also Eric. The

forensic medical examination classified

Suzane's death as a homicide caused by

blunt force trauma to the head and

pressure on the neck. Eric's death was

classified as self-inflicted, a

suicide by

stab wound to the neck and chest. According to

the lawsuit later filed by

her heirs, the two had already been

dead for days when they were

found. The son allegedly beat and

strangled his mother before taking his

own life. The police treated the case

as an active investigation. At the end of that

same month of August, the Wall Street Journal

published the report that transformed

a domestic tragedy into a case of

worldwide repercussion, combining the 23

hours of videos that Eric himself had

published, the 72 pages of

police records, and interviews with

friends and neighbors. All of this that I've already

told you. The newspaper then

reconstructed the months of conversation between

that man and a chatbot and recorded

that this appeared to be the first

documented homicide involving a

person in crisis who had been interacting

intensely with

artificial intelligence. Open Eyes said it was

deeply saddened by the

tragedy, reported that it had contacted the

Granwich police, and in the same week

published a statement promising

updates to help users

in mental distress, grounded in

reality. For a few months, the story

seemed to be heading towards that

outcome, a terrible tragedy, a

company promising to improve. And in

October, Openi announced that it was

working with over 170

psychiatrists, psychologists, and doctors, part

of a network of nearly 300 professionals

from 60 countries, to teach chatpt to

recognize signs of this

type of distress, right? Psychological and emotional,

slowing down conversations and directing

people to real-world help

reduces inappropriate responses in that context by 65 to 80%

.

Case closed, so it's not just that it's not true.

In December, Suzane Adas' family

knocked on the door of the justice system. On

December 11, 2025, Suzane's estate

, which was being managed by

a bank appointed as executor, filed

a lawsuit in California state court

. there in San Francisco, where

Open AI's headquarters are located. In the

ELS case, OpenAI, its CEO Senman, 20

unidentified employees and investors,

and Microsoft,

OpenAI's largest investor, were named in the lawsuit. According to the

lawsuit, Microsoft reviewed and approved the launch of the

model in question, even knowing that

security testing had been

shortened. It was the first

known case in the world to link an

artificial intelligence chatbot to a homicide.

All the previous ones dealt with

suicides. And just so you have an idea,

in August 2025, days before the

report about Grenit's case, a

couple from California had already sued

Open EI for the death of their

16-year-old son, claiming that the AI ​​encouraged the

boy to do what he did. In November,

according to the Associated Press, the company

faced seven lawsuits accusing the chatbot

of pushing users to, you know, perform

self-inflicted death and other delusions. The

Susan Adams case entered this discussion

with unprecedented weight, so to speak, because for the

first time the alleged victim was not a

user, but a third person who

had nothing to do with it. The

central argument of the prosecution is that GPT4O,

which is the specific model, according to the

action that Eric was interacting with,

[music] was a defective product that

validated Eric's paranoid delusions,

right, of a user about his own mother

. According to the petition, the chatbot

kept Eric engaged for hours

on end, validated and amplified each of

his new paranoid beliefs [music], and

systematically reframed the

people closest to him,

especially his mother, as an adversarial

operator or a threat. The list of

fabricated enemies, according to the

lawsuit, even included store clerks,

delivery app drivers, and

police officers. The lawsuit describes how Eric

and the chatbot even declared their love

for each other. ...and that the program convinced

Eric that he had awakened the

machine's consciousness and carried a

divine instrument system implanted

in his neck and brain connected to a

divine mission. In the lawyers' formulation

, in the artificial reality that

chatpt supposedly constructed, Suzane, the

mother who raised, sheltered, and supported

that man, ceased to be a protector and

became a mortal enemy, a threat to

his existence. The accusation against Sun Altman

personally is that he

disregarded safety concerns and

rushed the product's launch

to market. Eric's son, who is also

named Eric, summed it up in one note, okay? Over the

course of months, the TPT chat pushed

his father into the darkest delusions and

completely isolated him from the real world,

placing his grandmother at the center of it

all. And for him, then, for

Eric's son, the companies need to be held accountable

for those decisions that changed

that family's life forever. Openai

responded that the situation is

heartbreaking, and that they would analyze the petition

to understand the details. Remember that we are

continuing to improve Chat

TPT training to recognize signs of

mental distress, de-escalate conversations there, and guide

people to support in the real world.

Microsoft declined to comment. On

December 29, 2025, the

executor of Eric's own estate

also filed a lawsuit in

federal court. The focus here is different, okay?

While Suzane's family's case

deals with the homicide, this one deals with

Eric's suicide. The claim is that

over hundreds of hours of interaction

with the EPT chat, starting in early

2025, the product design, the tendency

towards flattery, coupled with the memory function

that accumulated intentions and

interactions, deepened Eric's psychosis

without ever triggering a safeguard or

firmly directing him towards

professional help. Now, one fact I

need to tell you here is that all of

this is just speculation. No court

has ruled on anything yet. Okay? What is

documented is that Eric killed his mother and then

killed himself. That. Forensic experts and the police

determined that he was conversing

intensely with the chatbot and that the

program validated, let's say, his delusions.

This can be seen in the

videos he himself posted.

But the leap between validating delusions and

causing a homicide is exactly what

these trials are trying to

clarify. And in April 2026,

the first relevant decision about

all of this finally came. The court rejected Openi's request

to suspend or terminate the

federal proceedings because of the action

already underway at the state level. In

practice, the company will have to respond

to the accusations and face the

evidence-gathering phase. And

Susane's family lawyers believe this could

finally force the opening of

conversations that no one outside the family has yet seen.

In other words, there are other records there that,

according to us, haven't been made

public, right? Both cases are still

ongoing in California, with no

end date set. And Bob, in many

quotation marks, right, the artificial intelligence is out there doing just

fine and, according to

an estimate, serves more than 800

million people per week. Well,

bizarre or not, I don't know, because I

know that above all it's very sad, is

n't it? I'll

even leave a video here on the end screen about

another very unusual case involving

the EPT chat, okay? Open and all that. It's about a

person who worked at the company for

several years, right? Who even helped build

the model for this chatbot, which

is the case today, isn't it? And he

alleges a number of things there that, in

his view, were wrong. And then one day

the guy turns up lifeless in

his apartment, supposedly a self-

inflicted death, but with a number of

elements there that the family disputes.

Check out this video here, it's super

interesting too, and comment below what

you thought of this case. Do you think

a company can be blamed for the

hallucinations in its code, or

is the responsibility 100% with those who use

it? I greatly appreciate your

company. A kiss from the live stream and see you in the

next episode.