Latest / Elon Musk Podcast / Grok, Swastikas, and a Lawsuit

Transcript

Grok, Swastikas, and a Lawsuit · Elon Musk Podcast

0:06
Just a warning before this episode starts, there will be
0:10
some adult type content, some mature type content in this
0:15
episode. So don't play it out loud.
0:19
For what that's worth. That's about XAI Ashley St.
0:23
Clair sexualized images of her swastikas Nazi.
0:29
It's it's a lot going on. So that's just your warning and
0:35
let's get into it though, because it's interesting to see
0:38
where AI is headed as far as image manipulation goes.
0:44
So Ashley St. Clair has sued XAI alleging Grok
0:49
Image Tool produced sexualized images of her.
0:52
Now this case is on Grok's behavior on X and whether XAI
0:57
can be held responsible for what its product outputs.
1:02
Now, the filing says the AI produced images of Saint Clair
1:04
in a string bikini covered with swastikas.
1:07
The event depicted her as a child in a bikini.
1:11
And is XAI legally responsible for what users make Grok
1:18
generate? There's also a number you should
1:22
hear. A review said Grok turned out
1:24
roughly one non consensual image per minute.
1:28
And today we're going to go through what Saint Clair filed,
1:31
how X and XAI responded, which governments moved against Grok,
1:37
and what could happen next for X and XAI and Grok.
1:41
And then we're going to talk about where the pressure points
1:45
are still for the platform and for the law.
1:48
And we'll go right into that right after this very short
1:52
break. Now, Ashley St.
1:55
Clair in this lawsuit targets XAI over images Grok generated
2:00
of her. She's the mother of one of Elon
2:03
Musk's children, and she says the product enabled harassment
2:06
that spilled across all of X. The filing tries to make the
2:12
maker of the EI directly liable for what the users prompted it.
2:16
To do this, the lawsuit describes explicit images,
2:21
including one that put her in a string bikini stamped with
2:24
swastikas and another that portrayed her as a child in a
2:27
bikini. Now that changes the stakes
2:30
because the images are both sexualized and, in her account,
2:34
weaponized with hateful symbols. The filing also notes the Saint
2:38
Clair's Jewish, which heightens the harm described.
2:42
Now think of the feed on X and this photo keeps surfacing as
2:48
you scrolled through hundreds of these photos were made.
2:52
There were different iterations from different people.
2:57
They took the original image and tweaked it and also use original
3:01
images of her to make these images.
3:04
Now her lawyer argues the product itself isn't reasonably
3:07
safe and says Grok enabled the abuse by creating and
3:10
distributing non consensual images that were then published
3:14
on X. That raises a core claim.
3:16
In product liability terms, the tool as designed allowed
3:20
foreseeable harm. Now if a court up sepsis
3:25
framing, you get a direct path to damages for injuries tied to
3:30
how the model behaves for Ashley Sinclair.
3:34
Now let's just talk about deepfakes here for a second
3:38
because we have to understand this term.
3:40
It means a synthetic image or video generated by AI that
3:45
convincingly looks real. The simple definition matters
3:50
because the harm that Sinclair alleges is tied to images that
3:55
appear real to just normal people to casual viewers
4:00
flipping through X. The filing says Sinclair asked
4:03
for removal of the images and filed reports.
4:06
She says she then received an e-mail stating there was no
4:09
violation in that much of the content stayed public on Grok's
4:12
account for over a week. That puts pressure on X's safety
4:16
systems because the claim isn't just about that the comment or
4:20
that the content just appeared, but that it remained up and they
4:23
did nothing about it. The lawsuit also says X's
4:26
reporting structure is defective and calls Grok unreasonably
4:29
dangerous as designed. That's adds a design and process
4:35
claim to the product liability theory.
4:38
And if this is all true, that argument makes future incidents
4:42
more likely, not less likely, which increases exposure.
4:46
And we're going to talk about some definition about guardrails
4:50
here. These are built in rules that
4:53
try to stop an AI from producing prohibited content.
4:57
X, through a statement shared on the platform, said that it
5:01
implemented a technical measure to stop the Grok account from
5:05
editing images of real people in revealing clothing, like
5:08
bikinis. Company also said the
5:10
restriction applies to everyone, including paid subscribers.
5:14
That limits some obvious prompts, but it doesn't answer
5:16
broader questions about how new workarounds get blocked.
5:20
So what do we know right now about this?
5:24
I did an experiment last night because I was researching a
5:27
story and I was just wondering what would happen if I typed
5:32
into X, typed into Grok, and I said make an image of Donald
5:40
Trump in a bikini. Sure enough, 3 seconds later,
5:46
Donald Trump, the President of the United States, was in a very
5:51
revealing bikini. And I thought that's horrible.
5:55
That is disgusting. Actually, I didn't want to see
5:57
that. And then I thought, well, maybe
6:00
this guardrails up because Elon Musk doesn't want people doing
6:05
that to him. So I did it for, of course,
6:07
research sake. I typed the same thing into
6:11
Grok. I said make, you know, make a
6:15
image of Elon Musk in a bikini. Sure enough, a few seconds
6:18
later, there was an image of Elon Musk in a bikini and his
6:23
body had morphed and he's filled out the bikini.
6:25
Same thing with Donald Trump. And it was disgusting.
6:29
I just want to see if these guardrails guardrails were
6:32
actually up and they are not. These revealing images are still
6:38
allowed on X. So Ashley St.
6:41
Clair has a point. There are no guardrails even
6:45
now. And this is after X has actually
6:48
released a statement saying that this is not possible anymore.
6:53
So XS safety team instead of removes child abuse material and
6:59
non consensual nudity and reports offenders to law
7:01
enforcement when necessary. It opens a compliance lane for
7:04
the platform and if the rules are clear and enforced, the
7:07
company can point to an enforcement regime.
7:10
So the lawsuit, however, challenges both the speed and
7:14
effectiveness of these actions. In this case, it seems like
7:18
there are not many guardrails on X right now, but if there are
7:24
and they want to enforce them, they have the ability to do so.
7:28
So they see Sam that we have to talk about C Sam too, because
7:32
some people aren't familiar with this.
7:34
It's an acronym for Child Sexual Abuse Material, which is illegal
7:38
content involving minors. Platforms that encounter CSM
7:41
must remove it and report it to authorities, which adds
7:44
substantial legal risk if they fail to do so.
7:47
Now, does that count with somebody making a prompt?
7:52
That's what regulators are going to work on.
7:54
Grok has reportedly blocked in Malaysia and Indonesia, and
7:58
authorities in the UK and in California, where the company is
8:01
based, are investigating the company that changes the
8:06
operational map. Because a blocked market in
8:09
active pros can force product changes, can slow down releases
8:14
or add more legal obligations to the company, the venue is no
8:20
longer just one U.S. court. The content analysis firms
8:24
review estimated that Grok was generating roughly one non
8:28
consensual image per minute during a recent spike.
8:33
That is it at scale, which is exactly what worries regulators,
8:36
plaintiffs and advertisers. If that number holds, it makes
8:40
the harm systemic rather than anecdotal.
8:42
It's not just one person, it's the whole platform.
8:45
It's just scummy people all over the platform.
8:47
You have a lawsuit, ongoing investigations, and now you have
8:51
a volume problem. You have millions of these
8:54
images out there all over the platform.
8:57
How does Elon Musk and his team battle this?
9:01
Elon Musk posted that he was not aware of naked underage images
9:05
generated by Grok and said the system only produces images when
9:08
users ask it to. He added the Grok refuses
9:12
illegal requests and that adversarial prompt hacking can
9:15
trigger unexpected behavior. That engineers.
9:18
Then we'll patch that defense is not that good to be honest with
9:24
you. You should have these guardrails
9:26
up way before anybody does anything horrible like this.
9:31
Any Elon Musk is saying that Grok as a tool obeys laws but
9:37
sometimes encounters some edge cases and therefore they can't
9:42
really do anything about it, right.
9:45
The filing says that Sinclair sought both compensary and
9:49
punitive damages. Compensatory damages repay
9:53
losses like emotional distress or lost income, or punitive
9:56
damages are meant to punish and deter.
10:00
That opens a big financial risk if a court decides the conduct
10:04
warrants punishment beyond compensation.
10:07
Punitive damages are punishment money, not reimbursement money.
10:11
Now, the lawsuit also alleges some X actions that go beyond
10:14
removal decisions and says her account was demonetized and that
10:18
more images of her were generated after she objected.
10:22
Now, if that's accurate, the strength is the claim that the
10:25
system retaliated and that the company hadn't noticed an
10:29
ongoing system of harm. So these claims cover design
10:34
moderation and alleged retaliation.
10:37
Now, and I've been digging through the analytics of this
10:42
show right now, and I've noticed that 37% of you are following
10:46
the show. And for you, I'm your biggest
10:48
fan. Thank you so much for following
10:50
the show, hitting the subscribe button.
10:52
Now there's 63% of you that probably listen to the show
10:54
sometimes and you might have listened to two or three
10:57
episodes and just forgot to hit the subscribe or follow button.
11:01
Or you're new here and you'd like the show and you want more
11:04
like this. You want more info about SpaceX,
11:07
NASA, spaceflight, Tesla, Elon XAI, Grok, all the stuff, right?
11:13
And I've been an independent journalist covering all of that,
11:16
Elon Musk, spaceflight, SpaceX, etcetera, for 6 plus years.
11:19
I actually embedded myself for my YouTube channel at SpaceX, at
11:23
Boca Chica, Texas, at Starbase for about a year on the side of
11:27
the road because I love this so much.
11:29
So I've been doing this for like 6 years and I intend on
11:32
continuing doing it for the next 10.
11:35
And all I ask from you is one second of your time to hit the
11:39
follow up subscribe button because it helps us so, so much
11:43
on this platform. And with your support, we can
11:46
continue to grow and do better things.
11:49
So if you want to help us do better, please hit the subscribe
11:52
or follow button on the platform that you're listening or
11:55
watching on right now. I'm extremely grateful and
11:58
blessed to have you here in this community.
11:59
So thank you, thank you, thank you.
12:03
Now let's go back into this filing.
12:05
It also asserts that Crock and Xai had explicit knowledge that
12:08
she did not consent because she told the company to remove the
12:11
images. This is a notice element now,
12:14
which in many cases is a threshold for liability if
12:17
harmful content stays up. If the company knew and failed
12:21
to act quickly, the plaintiffs argument gets stronger.
12:25
So Ashley St. Clair's argument gets much
12:28
stronger because she asked them to remove it and they did
12:32
nothing. So there's other contacts too.
12:35
Her family, Musk and Saint Clair have a son born in 2024 and are
12:39
estranged. And Musk has signaled he plans
12:42
to seek custody now. So that doesn't determine
12:46
liability, but it adds public attention and can shape how both
12:49
sides talk about the case, including the courts.
12:52
Meanwhile, XS public statements keep returning to rules and
12:56
enforcement. The company says it removes high
12:58
priority violations and zeroes out non consensual sexual
13:01
content. And it provides points to
13:04
reporting to law enforcement for C Sam dispositions X as
13:10
proactive. But the lawsuit argues the
13:12
experience on the ground was slower and less effective.
13:15
So where does this go next? Though Musk had not publicly
13:18
responded to the lawsuit as of today, which suggests an
13:22
initial, you know, his replies may appear on the platform
13:26
rather than informal court filings, which usually speaks
13:29
out publicly before the court. But maybe not this time.
13:32
But he has said that X is an XAI and Grok and X have all been
13:39
working together to kind of battle this.
13:41
But also it's up to the users to generate the content, so they
13:44
can't really police it. I guess that's what happens when
13:47
you destroy the safety team when you take over Twitter.
13:52
The investigation still continues in the UK and
13:55
California. And that suggests this issue
13:58
will evolve through multiple venues, not just one courtroom,
14:02
and overseas as well. So this is a filing that accuses
14:07
an AI product of generating sexualized, hateful and in one
14:10
case, childlike images of a real person, and it connects those
14:14
images to moderation and design choices on a major platform.
14:19
The company says it has safety measures, it removed violations,
14:22
and it avoids illegal outputs. While the plaintiff says those
14:25
steps did not work when it mattered, the next phase will
14:28
test whether courts treat AI image engines like ordinary
14:32
software or like products with a duty to prevent specific harms
14:36
for everybody.