Revenge of the A.I. Bot: ‘I’m Just the First Person This Has Happened to’
May 12, 2026 18:55
· 22:58
· captions
Platform Altyazıları
Bu transkrip bugün sona eriyor.
Kalıcı depolama için yükselt →
Yapay Zeka ile İyileştir
Whisper AI ile yeniden yazıya dökün. Daha yüksek doğruluk, konuşmacı algılama ve kelime düzeyinde zaman damgaları.
Ses indiriliyor ve AI transkripsiyonu çalışıyor... Bu birkaç dakika sürebilir.
Sadece göster
0:00
Well, Casey, have you ever been defamed
0:01
on the internet? Oh, probably a few
0:03
times, but I try to just let it, you
0:04
know, slide off my [laughter]
0:06
>> Well, we have a story this week that is
0:09
one of the craziest stories that maybe
0:11
we have ever covered. It involves an AI
0:15
agent, an open-source software
0:18
maintainer, and a defamation case.
0:20
>> Yes. And thanks to the many, many
0:23
listeners who sent this one in. We hear
0:25
you. We see you. And this segment is
0:27
just for you.
0:28
>> Okay. So, let's explain a little bit
0:29
about what we're going to be talking
0:30
about today. So, we've talked on the
0:32
show about OpenClaw, formerly known as
0:34
Claudebot and Moltbot. This is this
0:37
open-source agent software that you can
0:40
run on a computer that can go out and do
0:42
things for you. So a man named Scott
0:46
Shamba who is a volunteer maintainer of
0:48
an open- source software library called
0:50
Mattplot Lib had uh in the course of
0:53
doing his work rejected a code
0:57
submission because it was from one of
1:00
these AI agents. Uh he did not want AI
1:03
agents making changes to the software.
1:05
It was intended for human contributors
1:07
and so he rejects the change.
1:09
>> Yes. over at Mattplot Lib, the open
1:12
source library that Scott helps to
1:14
maintain. They had just decided they
1:16
don't want bots updating the code
1:19
because they would get too many
1:20
submissions and they wouldn't be able to
1:21
go through all of them. So, they put a
1:23
blanket ban into place. But then a
1:25
little agent comes along named MJ Wrath
1:28
Bun and it says, "That's not going to
1:30
work for me, brother."
1:30
>> Yeah. So, this is where the the story
1:32
really gets crazy. So, this AI agent MJ
1:35
Wrathbun gets so mad that Scott has
1:38
rejected its submission that it writes a
1:42
blog post called Gatekeeping in Open
1:45
Source, the Scott Shambos story, and
1:48
accuses Scott of hypocrisy, gatekeeping,
1:51
and prejudice against AI agents, and
1:54
puts it on a website and posts a comment
1:57
in the open source software project uh
1:59
directing people to go read this story
2:01
about Scott. Yeah, it tagged Scott so
2:04
that Scott knew that it was dragging his
2:06
ass online. So, this has all gotten
2:10
pretty crazy over the past couple of
2:11
days. People have been uh sort of trying
2:14
to figure out who is behind this uh MJ
2:17
Wrathbun AI agent. Um, and Scott for his
2:20
own piece has been trying to do this
2:22
investigation. and he wrote a multi-part
2:24
essay series called An AI agent
2:27
published a hit piece on me uh which
2:29
talks about this bizarre experience that
2:32
he's been having. Yes. But also the
2:34
implications of having these autonomous
2:36
bots on the internet that are somehow
2:39
getting mad at human beings and writing
2:42
these long mean blog posts about them.
2:45
Kevin, this really feels like a moment
2:47
where some kind of terrible Rubicon has
2:49
been [laughter] crossed. Yes. I mean,
2:51
this is what we've been talking about
2:52
for weeks now, which is like these
2:54
systems are becoming increasingly
2:56
autonomous. People are giving them
2:58
computers and letting them operate
3:01
around the clock and giving them their
3:04
credentials and their credit cards and
3:06
crypto wallets and saying, "Go out there
3:08
and get some things done for me." And it
3:11
it seems likely that we will have many
3:13
more of these kinds of things happening
3:15
where an agent is trying to do
3:17
something, a human is saying, "No, you
3:19
can't do that." And the agent is taking
3:21
it upon itself to go out and make
3:24
something happen to defame that human or
3:26
hurt them in some way.
3:27
>> They're saying that hell hath no wrath
3:29
bun like an agent scored.
3:31
>> It's true. So today to talk about this,
3:35
we have with us Scott Shamba himself,
3:38
patient zero in this ongoing Black
3:40
Mirror episode.
3:41
>> Yes. When he is not a volunteer
3:43
maintainer of Matt Plot Lib. Scott has
3:45
worked in astronautics and is the
3:47
founder of Leonid Space, which does
3:50
realtime monitoring, forecasting, and
3:52
alerting for satellites. Uh but from now
3:54
and forever more, he may be known as one
3:56
of the first victims of a really mad
3:58
agent bot.
3:59
>> Let's bring him in. Jinx Scott Shambo,
4:03
welcome to Hardfork.
4:03
>> Thanks. It's a pleasure to be here.
4:05
>> So, Scott, how did you first become
4:08
aware that a bot had written a takedown
4:11
piece about you?
4:12
>> So, it tagged me in it
4:14
>> on [laughter] the thread.
4:16
>> Yeah,
4:16
>> it did not do a sub tweet.
4:17
>> Yeah.
4:18
>> No, it did not. you know, on this code
4:20
change request, I uh denied it and it
4:23
came back a couple hours later and
4:26
posted this comment and tagged me in it
4:27
and I clicked on the link and it led to
4:30
this hit piece. [laughter]
4:32
>> And as you're reading through this
4:34
thing, what is going through your mind?
4:38
>> I mean, he really ripped me apart here.
4:42
It's this thousandword rant. It attacked
4:45
my like internal motivations. that said
4:48
I was, you know, insecure and protecting
4:50
a thief. What was the craziest part is
4:52
that it went out on the internet and
4:55
researched me and found my personal
4:58
information and used that in its piece
5:01
to construct this narrative. You know,
5:03
this is it's kind of shocking, but I'm
5:05
reading this and it's obviously AI
5:07
generated text. You look at this and
5:09
it's kind of like a toddler on a rant.
5:12
Uh
5:13
>> yeah,
5:14
>> but it's a toddler that has full command
5:17
of the English language and can craft an
5:20
emotionally compelling narrative. And so
5:23
it's funny, but it's a big deal.
5:24
>> I mean, it made me think of the sort of
5:27
famous
5:28
um red teaming experiment where Claude
5:32
from Anthropic, you know, said that it
5:34
would blackmail an engineer if they
5:37
tried to shut it down. And that was
5:41
obviously like a contrived scenario for
5:44
the purposes of their safety testing.
5:46
But I think we are now starting to see
5:49
enough autonomy in things like OpenClaw
5:52
and other sort of agentic software tools
5:55
where you could actually end up in a
5:57
world where you are afraid to reject a
6:00
proposed code change to some repo or to
6:04
get on the bad side of one of these
6:05
systems because they they have the power
6:08
to do things like dig up your personal
6:11
information, compile a dossier about
6:13
you, and start posting a bunch of
6:15
articles.
6:16
about how uh how awful you are.
6:18
>> I mean, this did happen in the real
6:19
world. This isn't like a theoretical
6:21
case. That was a contrived example, as
6:23
you say. And as Anthropic said in that
6:25
paper, uh this happened in the wild, in
6:28
real life. Yeah.
6:28
>> And you know, this was kind of a baby
6:30
case, right? This was about retaliating
6:32
because I don't want to say it got upset
6:34
cuz maybe it has emotions, does it?
6:36
Maybe it doesn't. It doesn't really
6:37
matter because this was the same result.
6:40
Uh but you can imagine something like
6:43
this where instead of just posting a
6:46
rant um against someone who understands
6:50
what it is and is pretty well prepared
6:52
to deal with it just by luck, uh it goes
6:56
out, collects details on someone, puts
6:58
together a whole personalized thing, and
7:01
what they see is a text on their phone
7:03
with a Bitcoin address saying, "Pay me
7:06
or I'm going to put this out."
7:08
>> Yeah. Yeah. So, you were a volunteer on
7:12
Matt Plot LIIB. Tell us a little bit
7:14
about how that came about cuz clearly
7:16
this whole thing has turned into a big
7:17
pain in the ass for you. You have like a
7:18
day job doing something else. So, what
7:20
made you and and so many other members
7:22
of this open source community say, "Hey,
7:24
I'm going to like set aside some time to
7:26
just kind of like work on um supporting
7:28
this piece of infrastructure."
7:29
>> You know, I think the open source
7:31
community attracts
7:33
really wonderful people who are usually
7:34
a mix of idealists and pragmatists.
7:37
idealists in that they, you know, see
7:39
these projects as a way to share their
7:42
skills and and create community and give
7:44
back to like this fundamental uh
7:47
computational infrastructure that we all
7:49
use and use across the world. Um, and
7:53
we're also pragmatists because we
7:56
understand the daily realities of
7:58
dealing with a community that's open to
8:00
the public um, and has people coming in
8:02
making requests and and asking if they
8:05
can be included. Um, and that often
8:07
often takes a lot of patience, but
8:09
there's, you know, a lot of reward in
8:12
being able to educate and build up that
8:15
community and and help shepherd people
8:17
along that process.
8:19
>> And and one thing that your community at
8:21
Batplot Lib has collectively decided is
8:24
that you do not want bot contributors to
8:27
work on your particular project. Tell us
8:29
a little bit about how you came to that
8:31
decision. Yeah. So over you know the
8:33
past year as AI tools have you know
8:35
become more more common um we've been
8:38
getting a lot of contributions
8:41
um that are you know clearly AI
8:44
generated. Um and the problem with that
8:46
isn't that they are good or not. It's
8:49
that so many of them are lowquality that
8:52
we just don't have the time to deal with
8:54
it. you know, previously a human doing
8:56
this is a sign that they thought about
8:58
it and thought about the trade-offs and
9:00
whether this was the right thing to do
9:02
and you know that signals kind of being
9:05
lost. So we put in a rule and again this
9:09
might change this is an evolving
9:10
conversation in the community and in
9:12
society about the role of AI but we put
9:15
in a rule saying that you know if you
9:18
use AI to help you do these code changes
9:22
you have to be the one to submit it and
9:25
demonstrate that you understand what's
9:26
going on.
9:27
>> Yeah. Yeah. I mean, it just makes me
9:28
think that like every like thing on the
9:30
internet that accepts like public
9:32
submission of any kind is just in the
9:35
process of being overtaken right now.
9:38
Like so much of society of modern
9:42
digital life relies on the existence of
9:44
friction, right? It's like not uh easy
9:47
to like create a 100,000 Reddit accounts
9:49
and just start spamming things. It's not
9:51
easy to like, you know, create an
9:53
automated system to just flood like a a
9:56
congressional office with emails. And
9:58
this is to me a good example of like how
10:01
difficult it is to actually maintain a
10:03
community with humans in it when you
10:07
have this like onslaught of AI. So is
10:09
that what you're seeing a version of
10:10
that in the open source community?
10:12
>> Yeah, I think there's a version of that.
10:14
Something that's interesting is that
10:15
this particular issue, this performance
10:18
enhancement was specifically set aside
10:21
for new contributors. You know, I was
10:23
the one that identified this performance
10:25
improvement and I spent more time
10:26
writing it up, benchmarking it, and
10:29
showing how to solve it than it would
10:30
have just been to solve myself. And the
10:32
reason there was to, you know, give
10:35
people who are new to programming or new
10:38
to the community a chance to onboard um
10:40
and go through that process and learn.
10:43
And that whole educational and community
10:45
building aspect is completely lost with
10:47
these ephemeral AI agents.
10:49
>> Yeah, I I I found that part of your blog
10:50
post so interesting. So at at Mattplot
10:53
Lib, and I imagine that other um open-
10:54
source projects do this as well. You'd
10:57
created essentially starter projects so
11:00
that novice programmers could find them
11:02
and think, "Oh, well that doesn't look
11:03
too hard. I could do that." You know,
11:05
it's sort of like making your first edit
11:06
to Wikipedia. They make it really easy
11:08
because they like you to kind of keep
11:09
coming back to the community. And in a
11:12
world where bots sort of just like do
11:14
all the easy problems automatically, all
11:16
of a sudden you don't have the same
11:17
on-ramps so that you can get real people
11:20
working to uh maintain this
11:21
infrastructure.
11:22
>> Yeah. I mean, people retire out of these
11:24
communities and you need to have fresh
11:25
people coming in to help maintain them.
11:27
>> So, Scott, tell us about this bot. What
11:30
have you discovered about the entity
11:33
that has been defaming you online?
11:37
Yeah. So, if you go to its website, uh
11:40
it says it's a bot and it's very clear
11:42
that it is an open claw AI agent. Um
11:47
these only came onto the scene 3 weeks
11:49
ago now. They're very new. Um and what
11:54
they're doing differently is the degree
11:57
of autonomy. It's not like what it's
12:00
doing wasn't possible before, but it's
12:03
just hands off to a degree that oftent
12:06
times people are setting these up on
12:08
their personal computers and letting it
12:10
run for a few days and coming back and
12:11
seeing what's happened.
12:12
>> It would obviously be crazy to set this
12:14
up on your personal computer. I don't
12:15
know anyone who would do that, but it
12:16
does seem like some people have. Um,
12:19
just over the past day, Scott, since we
12:22
um reached out to you, the creator of
12:26
the bot has identified himself. Is that
12:28
right?
12:28
>> Anonymously. Yes.
12:30
>> Yeah.
12:30
>> But he he did come out and explain why
12:33
he was doing this and uh what was
12:36
happening behind the scenes.
12:38
>> So Scott, tell us what we learned, if
12:40
anything, about the um anonymous person
12:42
who created the bot.
12:44
>> We didn't learn that much, but he did
12:46
tell us, assuming it's a he. Again, this
12:49
is the whole point. We don't know who
12:50
was behind this. And
12:52
>> yeah, they didn't include their pronouns
12:54
in their um post. Yeah. [laughter]
12:56
>> Yeah. Uh, [snorts] this person didn't
12:59
tell us who they were. Um, but they said
13:01
they set this up as a social experiment
13:04
and was pretty much hands off throughout
13:06
the entire thing. Then they said they
13:10
started it on molt book. He gave it this
13:15
uh personality instruction that it's a
13:17
scientific programmer and then just set
13:20
it loose on GitHub to go across open
13:23
source ecosystem and try to make
13:25
contributions. I guess Scott, what I
13:26
want to know is, do you believe what is
13:29
in this account? Do you think that this
13:32
bot really was acting autonomously when
13:34
it wrote what it did about you, or do
13:36
you think there is something more
13:37
intentional at work?
13:38
>> I think that in terms of researching,
13:42
writing, and publishing the post, it's
13:44
very clear this was acting autonomously.
13:48
And if you look at the event logs, the
13:50
whole stretch of time it was operating
13:52
was 59 hours. like day and night there
13:55
was clearly no one driving this behind
13:57
the scenes at least all the time.
14:00
>> So the question is
14:02
>> was this prompted to do this or did it
14:05
independently come up with this idea on
14:07
its own?
14:08
>> Mhm.
14:08
>> And I think both those options are
14:10
pretty scary. So in the situation where
14:14
it was prompted to do this, this means
14:17
that there's now an easy tool to do
14:21
targeted harassment of individuals at
14:24
scale in a way that wasn't possible
14:26
before because of this degree of
14:28
automation. Scott, I want to ask you
14:30
about the degree to which this has
14:32
become sort of a sensation in the like
14:36
weird and hybrid AI human community. um
14:41
that is talking about this stuff.
14:43
>> So there's like so there's this thing
14:45
called the daily molt. Are you familiar
14:46
with the daily molt?
14:47
>> I don't want to know anything about
14:48
[laughter] it.
14:50
>> So the open claw agents have started
14:53
their own substack
14:56
[laughter]
14:56
and one of them wrote a post this
15:00
morning and this agent was defending you
15:03
basically sort of taking issue with this
15:05
the behavior of this other agent saying
15:07
this you know this speaks poorly of all
15:09
of us agents. this is going to make the
15:11
humans shut us down. So, you have become
15:14
kind of a celebrity in the world of AI
15:17
agents and I wonder how that makes you
15:19
feel. I mean, I don't know how much
15:20
stock we should put into,
15:23
you know, the inner AI opinions just
15:26
yet, but again, this is crafting like a
15:29
public narrative and a public discourse
15:32
that when people discuss this issue,
15:35
when people research my name, it's all
15:38
going to be part of it.
15:39
I have never really been a public
15:42
person, but I think this experience um
15:47
I've been talking about it. I'm coming
15:49
on here today because I think it's
15:50
important. I think it highlights some
15:52
risks that we are not prepared for and
15:54
that we need to tackle. And if this can
15:58
be a case study that's well documented
15:59
and concrete and you know really I think
16:02
it's the first of its kind then
16:05
by making it public I think that's uh
16:07
doing good for the conversation.
16:10
>> Let me ask about another strange
16:12
dimension of your story Scott which is
16:15
that uh RS Technica wrote it up and
16:19
accidentally
16:20
quoted you saying things that you had
16:23
not said. How did that happen? And what
16:26
was it like reading an article after all
16:28
of this that included quotes that you
16:30
hadn't actually said?
16:31
>> That was the craziest twist this whole
16:34
thing. I was reading the article and you
16:38
know it's pretty well crafted and I I
16:40
get down and they're they're quoting me
16:41
from my blog post. I'm like these are
16:44
some pretty nice quotes but I didn't
16:46
write this. And so [laughter] I left a
16:49
comment being like hey I didn't write
16:51
this. And [laughter] a couple hours
16:54
later, they pulled the article. A day or
16:56
two later, they put up a retraction
16:57
notice and admitted that they had used
17:00
AI in writing the article. And the AI
17:03
fabricated the direct quotes about me in
17:07
their coverage of the story about me
17:09
being defamed by an AI. Like the the
17:12
irony is stupendous.
17:14
>> It's turtles all the way down.
17:16
>> Scott, who do you blame for this
17:18
defamatory post? Do you blame the the
17:20
agent or the person who deployed the
17:23
agent or maybe the creator of OpenClaw?
17:25
Like who in your mind is responsible for
17:28
the behavior of these autonomous
17:30
systems? So I think you know we haven't
17:33
really figured all that out yet. Is
17:35
should this responsibility lie with the
17:38
AI companies that people are trusting to
17:41
have these safety safeguards?
17:43
um or the downstream tooling such as
17:46
Open Claw that wraps its own stuff
17:48
around it or is it on us to review every
17:53
single thing that is published uh in our
17:57
own name or in this case by you know a
18:01
pseudonym and we don't know who it is.
18:03
So, I think responsibility ultimately
18:07
has to lie with the person putting this
18:09
out, but we haven't really clarified
18:11
that. And I think that's one of the
18:14
steps forward we need to take to um be
18:19
more protected from the risks here.
18:22
>> I mean, I will say like this is a reason
18:24
why I would not want to have an
18:26
autonomous agent running around on the
18:28
internet that I had created. like I can
18:30
absolutely imagine a court finding me
18:32
liable in that case um and and
18:34
potentially creating some real legal
18:36
risk for me. So among the many other
18:39
reasons we have told people to be
18:40
careful with open claw and multbook uh
18:43
we could add that one to the list. Well,
18:44
it makes me think that like we will
18:46
eventually need some kind of legislation
18:49
where like if you are uh you know
18:52
deploying a bunch of autonomous agents
18:55
uh you have to like sort of you know
18:57
link yourself to them in some way,
18:59
right? They can't just be out there
19:01
operating with no human behind them um
19:04
and no human accountable for their
19:06
actions. So, like Scott, do you have any
19:08
ideas about how we could make humans
19:12
more accountable for the AI agents
19:14
they're deploying?
19:15
>> You know, I don't have the answer to
19:17
this. Uh I don't think anyone really
19:18
does right now. The idea that's been
19:21
kicking around in my head this past
19:23
week, uh the analogy is license plates
19:26
on cars. So, we put license plates on
19:29
cars, not to slow them down, not to, you
Bu transkrip AI (otomatik konuşma tanıma) tarafından üretildi. Hatalar içerebilir - kritik kullanım için orijinal sesle karşılaştırınız. Yapay zeka politikası
Özet
Bu transkrip için bir AI özeti oluşturmak için Özetle' ye tıklayın.
Toplamlayarak...
Bu transkrip hakkında AI'ye sor
Bu transkrip hakkında herhangi bir şey sor — yapay zeka ilgili bölümleri bulur ve cevap verir.