WEBVTT

00:00:00.000 --> 00:00:02.077
Speaking of friction, let's talk about

00:00:02.160 --> 00:00:05.157
your refusal to allow the military to

00:00:05.240 --> 00:00:09.157
remove safety guardrails from Claude.
I don't want to get into the extensive

00:00:09.240 --> 00:00:11.357
details and back and forth here because

00:00:11.440 --> 00:00:16.357
other journalists are covering this on
a daily basis. But what was the crux of

00:00:16.440 --> 00:00:21.037
that decision? >> Yeah,
so I would I would say on DOD or

00:00:21.120 --> 00:00:25.597
DOW, you know, to be honest, we actually
had worked with them, right? So, you

00:00:25.680 --> 00:00:28.797
know, we we had provided our models
and you know, I'm I I'm actually believer

00:00:28.880 --> 00:00:32.477
and Danielle is a believer that we we
do need to defend our country, but you

00:00:32.560 --> 00:00:35.837
know, we we felt it's not worth
defending the country if we do things

00:00:35.920 --> 00:00:38.797
that go against
the values of this country.

00:00:38.880 --> 00:00:42.757
And there were two use cases that we
were uncomfortable with that we felt

00:00:42.840 --> 00:00:45.157
went against the values of this country.

00:00:45.240 --> 00:00:49.597
And and that was fully autonomous
weapons. So, you can imagine like a

00:00:49.680 --> 00:00:53.357
drone army that you know, one person has
their finger on the button and you don't

00:00:53.440 --> 00:00:54.917
have human soldiers. You have a you

00:00:55.000 --> 00:00:57.157
know, you have a drone army and you have

00:00:57.240 --> 00:01:02.677
AI powering this. And then and then uh
domestic mass surveillance, which is,

00:01:02.760 --> 00:01:05.717
you know, spying you know, using
the power of the government to spy on

00:01:05.800 --> 00:01:08.157
Americans. And and we thought those were

00:01:08.240 --> 00:01:11.997
pretty reasonable things not to allow.

00:01:12.080 --> 00:01:15.437
Uh and you know, unfortunately,

00:01:15.520 --> 00:01:17.277
the Pentagon

00:01:17.360 --> 00:01:22.517
uh didn't feel that way. And and so we
just we just couldn't agree to that. All

00:01:22.600 --> 00:01:26.637
the other companies have
agreed to that now, but I

00:01:26.720 --> 00:01:31.517
you know, I I we we all met
as co-founders. We said, you know, holy

00:01:31.600 --> 00:01:36.277
this could be really bad for the company,
but you know, we can't do this, right?

00:01:36.360 --> 00:01:38.917
>> agreed. >> Everyone
ever we we just we had you

00:01:39.000 --> 00:01:41.557
know, we had a meeting around a table
like this and you know, we just said

00:01:41.640 --> 00:01:46.477
>> Was that a dark night of the soul
kind of moment? What what what what was

00:01:46.560 --> 00:01:50.077
everybody willing to stick to those
principles or did some people have to be

00:01:50.160 --> 00:01:53.877
convinced or did you know going
in that this is this is the end of

00:01:53.960 --> 00:01:56.997
>> The co-founders believed this
unanimously. Like there there was a

00:01:57.080 --> 00:02:00.517
sense of like of like,
you know, well, you know,

00:02:00.600 --> 00:02:03.277
all right. I wish they hadn't done this,
but if they're going to do it, bring it

00:02:03.360 --> 00:02:05.677
on. >> Weren't you scared
of what would happen to you?

00:02:05.760 --> 00:02:09.357
>> Yes. >> [laughter] >> Yes. >> Were
you scared of what would happen to you?

00:02:09.440 --> 00:02:13.757
>> Yes. I mean, I think I think the thing
that It was It was a hard time. I mean,

00:02:13.840 --> 00:02:15.997
I don't think we want to sugarcoat that.

00:02:16.080 --> 00:02:19.397
It was a stressful time at the company,
but I think a thing that Dario

00:02:19.480 --> 00:02:21.157
>> it could be the end of the company?

00:02:21.240 --> 00:02:24.517
>> I think it crossed our minds.
Yeah. I think um

00:02:24.600 --> 00:02:28.917
I think Dario and I
and the co-founders though,

00:02:29.000 --> 00:02:32.877
the thing that we all had in common
was we just felt like this was the right

00:02:32.960 --> 00:02:34.997
thing to do. >> You were
willing for it to be the end of

00:02:35.080 --> 00:02:38.037
the company to stand
up to your principles.

00:02:38.120 --> 00:02:41.837
>> I think the reason
we founded Anthropic was

00:02:41.920 --> 00:02:43.877
our commitment to our values and the

00:02:43.960 --> 00:02:47.397
ethics that we chose to work with.

00:02:47.480 --> 00:02:52.757
>> And it felt like in the process
of sacrificing those values, we wouldn't

00:02:52.840 --> 00:02:55.797
have been the company that we
had started, if that makes sense.

00:02:55.880 --> 00:02:57.757
>> It It It you know,
the the way it happened

00:02:57.840 --> 00:03:00.717
>> What did you think would be the So I'm
sure if you're sitting around the table,

00:03:00.800 --> 00:03:04.517
you're looking at the best-case scenario
and the worst-case scenario. And in that

00:03:04.600 --> 00:03:07.717
moment, it doesn't look like there
is a best-case scenario. So what was the

00:03:07.800 --> 00:03:10.837
worst-case scenario?
>> At various times, they had made claims

00:03:10.920 --> 00:03:14.437
that they could stop other companies
from doing business with us, right? Not

00:03:14.520 --> 00:03:18.637
just stop us from working with the
government or being part of other

00:03:18.720 --> 00:03:21.157
companies' contracts with the government,
but that they could stop

00:03:21.240 --> 00:03:23.477
other companies from doing
business with us entirely.

00:03:23.560 --> 00:03:25.677
>> Period. So you're not >> Period.
Not only you're not doing

00:03:25.760 --> 00:03:29.077
business with the government, you're not
doing business with anybody Anybody else

00:03:29.160 --> 00:03:31.797
that does business with you.
>> Yeah, or you can't do business with

00:03:31.880 --> 00:03:34.757
anyone that does business with the
government in any way, which ends up

00:03:34.840 --> 00:03:40.237
being everyone or everybody, right?
Um I think the attitude we had was

00:03:40.320 --> 00:03:41.597
look, this is really screwed up.

00:03:41.680 --> 00:03:45.917
Everyone knows this is wrong. Lots
of people are going to be on our side.

00:03:46.000 --> 00:03:49.877
We think this is going to work out okay,
but man, are we not sure. You know,

00:03:49.960 --> 00:03:53.637
there were moments there where it
was like, oh man, you know, are are we

00:03:53.720 --> 00:03:55.717
wrong? Is this going to be
the end of the company? Like

00:03:55.800 --> 00:03:57.077
>> What were you thinking, Daniela?

00:03:57.160 --> 00:04:00.837
>> You know, I just keep
coming back to the co-founders

00:04:00.920 --> 00:04:03.197
felt like it was just the right thing to

00:04:03.280 --> 00:04:08.797
do. I think Anthropic's view has always
been that we are about policy and not

00:04:08.880 --> 00:04:11.877
politics. >> Yeah.
>> And Anthropic has leaned in, I think,

00:04:11.960 --> 00:04:16.237
very early before we even had a product
to the fact that, you know, we think

00:04:16.320 --> 00:04:21.477
that the role of government in the
development and thoughtful regulation of

00:04:21.560 --> 00:04:24.557
artificial intelligence is going
to be incredibly important. I think AI

00:04:24.640 --> 00:04:27.157
companies will have to work with
the federal government, with state

00:04:27.240 --> 00:04:31.917
governments, internationally, with with
NGOs. And I think our view was always,

00:04:32.000 --> 00:04:35.357
look, this is really about the substance.
This is not a political

00:04:35.440 --> 00:04:37.357
disagreement. This is a policy

00:04:37.440 --> 00:04:42.717
disagreement. And I think in some ways
that in it's not to say it was easy, but

00:04:42.800 --> 00:04:46.317
I think it made our decision easier
because we felt strongly, look, this is

00:04:46.400 --> 00:04:49.997
the right thing to do. These are the right
red lines. To Dario's point, we've

00:04:50.080 --> 00:04:54.237
we were actually the first AI company
to work closely with national security

00:04:54.320 --> 00:04:57.837
because we felt that that was in line
and in keeping with our values, except

00:04:57.920 --> 00:05:01.477
in these two areas that we felt
could have the potential to compromise

00:05:01.560 --> 00:05:04.237
democratic values. >> It's so interesting

00:05:04.320 --> 00:05:07.397
because most people don't get a decision

00:05:07.480 --> 00:05:10.317
at that scale that they have to make

00:05:10.400 --> 00:05:12.557
that they have to rely on their

00:05:12.640 --> 00:05:16.877
principles and their values. Um and and

00:05:16.960 --> 00:05:21.837
you're literally risking
a multi-multi billion-dollar company.

00:05:21.920 --> 00:05:23.597
But it all goes back to how you were

00:05:23.680 --> 00:05:26.477
raised and what how those principles

00:05:26.560 --> 00:05:31.157
were formed in you before you
had a big decision like that.

00:05:31.240 --> 00:05:34.717
>> know, what we talked about before we
ever started the company, what's it you

00:05:34.800 --> 00:05:38.517
know, would our past selves be proud
of ourselves if we if we gave in here, if

00:05:38.600 --> 00:05:43.077
we compromised? Like I I you know, I I
I that that made the decision very clear,

00:05:43.160 --> 00:05:45.157
right? >> And I I will
say, I mean, it wasn't

00:05:45.240 --> 00:05:48.757
nearly as, you know, publicly covered,

00:05:48.840 --> 00:05:53.517
but I think Anthropic has had, you know,
smaller decisions along the way that

00:05:53.600 --> 00:05:55.197
felt like they they laddered up to this.

00:05:55.280 --> 00:05:57.797
I mean, >> Yeah. >> now it
looks so obvious in retrospect.

00:05:57.880 --> 00:06:01.797
>> that because because to get to a big
decision like this in a moment like that

00:06:01.880 --> 00:06:06.877
where you have everyone around the table
and have to it means the that that life

00:06:06.960 --> 00:06:10.277
the life force is the way the universe
works is that you would have had other

00:06:10.360 --> 00:06:14.397
decisions other decisions other
decisions that laddered to that moment.

00:06:14.480 --> 00:06:16.757
>> Yes. >> Yeah. >> I think
there are probably three I can

00:06:16.840 --> 00:06:18.197
think of off the top of my head.

00:06:18.280 --> 00:06:20.757
>> ask you that. >> Dario
might have others as well, but I

00:06:20.840 --> 00:06:23.037
think, you know, one which, you know,

00:06:23.120 --> 00:06:26.277
now feels sort of very obvious, but was

00:06:26.360 --> 00:06:30.637
a bigger discussion at the time was the no
ads decision. So, as you can imagine,

00:06:30.720 --> 00:06:35.117
I think for the co-founders, for me and
Dario, it was not controversial at all.

00:06:35.200 --> 00:06:39.317
We said, "Look, we think the incentive
the incentives are wrong. AI is

00:06:39.400 --> 00:06:43.717
different. People have very private
conversations with these AI models.

00:06:43.800 --> 00:06:45.637
People upload their health information.

00:06:45.720 --> 00:06:47.997
They ask questions about their children.

00:06:48.080 --> 00:06:49.877
They ask >> Certainly finances.

00:06:49.960 --> 00:06:53.797
>> Finances. They talk about, you know,
>> Medical information. Medical finances,

00:06:53.880 --> 00:06:55.837
yeah. >> Challenges they
might be having with an

00:06:55.920 --> 00:06:59.037
important person in their life.
It's I think we

00:06:59.120 --> 00:07:04.437
want to treat our customers' data with
the utmost privacy and respect. And so,

00:07:04.520 --> 00:07:07.437
for the co-founders, it was never
a question. We said, "Look, we don't want

00:07:07.520 --> 00:07:11.517
to do ads, but there were quiet forces
that, you know, within the company or

00:07:11.600 --> 00:07:14.877
investors who were like, "Hey, this is
a great source of revenue. This is how

00:07:14.960 --> 00:07:18.877
Google and and Facebook and basically
every company that has a big user base,

00:07:18.960 --> 00:07:20.037
you know, that's just how it's done.

00:07:20.120 --> 00:07:23.717
That's how you make money." And we just
said, "Like, sorry, that's just not how

00:07:23.800 --> 00:07:25.597
we're going to do things.
We'll find another way."

00:07:25.680 --> 00:07:27.397
>> That was That was one
of the big ones, okay.

00:07:27.480 --> 00:07:28.837
>> Yes, that was that
was one of the big ones.

00:07:28.920 --> 00:07:31.717
>> You have one? >> I can think
of a couple. One is, uh, you

00:07:31.800 --> 00:07:35.397
know, there was this time,
I think, in 2025,

00:07:35.480 --> 00:07:37.717
um, you know, probably the whole year of

00:07:37.800 --> 00:07:42.157
2025, the sentiment in tech was, "Let's

00:07:42.240 --> 00:07:46.157
go ahead, right? Let's Let's go ahead
with the technology like, you know,

00:07:46.240 --> 00:07:49.437
worrying about risks, you know, they're
all kind of politicized language about

00:07:49.520 --> 00:07:51.157
why it's bad to slow down and be

00:07:51.240 --> 00:07:54.717
careful. And so, there was a moment in

00:07:54.800 --> 00:07:58.277
the middle of 2025

00:07:58.360 --> 00:08:01.957
where, you know, there was a bill being
considered in Congress that would have

00:08:02.040 --> 00:08:06.557
banned all regulation of AI by states
and wouldn't have put in any federal

00:08:06.640 --> 00:08:09.917
regulation of AI either. So,
basically it would have banned

00:08:10.000 --> 00:08:11.597
all regulation of the technology.

00:08:11.680 --> 00:08:12.517
>> I remember hearing about this.

00:08:12.600 --> 00:08:16.557
>> And And we, you know, we were trying
to decide what to do. We definitely didn't

00:08:16.640 --> 00:08:20.797
support it, but the question was, do we
just sit there and say nothing? Or do we

00:08:20.880 --> 00:08:24.517
speak out against it? And all kinds
of people said to us, "No, it'll be

00:08:24.600 --> 00:08:27.357
terrible. You'll break all your
political relationships. All the other

00:08:27.440 --> 00:08:31.117
companies will hate you if you speak
out against it." Um and actually some of

00:08:31.200 --> 00:08:33.917
that did happen when we
we spoke out against it. Um

00:08:34.000 --> 00:08:38.797
uh but, you know, we decided that this
was an area that we had to speak up,

00:08:38.880 --> 00:08:42.277
that was central to the values of the
company. So, I wrote an op-ed in The New

00:08:42.360 --> 00:08:46.037
York Times saying this is a bad idea. We
should You know, that's about as loud as

00:08:46.120 --> 00:08:51.127
you can be. And that bill got
voted down in the Senate 99 to 1.

00:08:51.210 --> 00:08:55.397
>> [laughter] >> Yeah. >> So,
you know, we we were clearly on the

00:08:55.480 --> 00:08:59.720
right side.
Clearly on the right side here.