The AI crisis is a leadership crisis

What should we make of the claim that AI has a greater than 10 percent chance of killing all humans within the next decade? 10 percent is no joke. For example, two weeks before the 2016 election, The New York Times gave Donald Trump just a 7 percent chance of winning. 

It's worth repeating: 10 percent is no joke. Greater than 10 percent is greater than no joke.

This isn't coming from some random user on X. It's coming from a senior safety researcher at Anthropic, the leader of the leading frontier AI companies. So, of course, AI is wildly unpopular with the very humans whose lives it might incinerate. 

According to a recent Gallup poll, only 27% of Americans trust businesses to use AI responsibly. Since that poll was published, given the growing number of reported incidents inside these frontier AI companies, what are the odds that public trust has improved?

In response, politicians have released statements, demanded hearings, promoted more regulation, and even called for stopping AI development altogether. Of course, this does nothing to ease anyone's concerns. Congressional approval is lower than AI, at 10 percent, which feels 10 percent too high, honestly.

While President Trump is more popular than AI or Congress, he doesn't beat them by much. He's currently sitting at a 32% approval rating. This week he got the leaders of these companies to come to the White House, sign a document, and stand awkwardly in front of the press. He also rebranded AI as Super Intelligence, hoping America's love for superhero movies will change how it feels about things.

It's worth noting where we are right now. While the purveyors of this technology build the most valuable companies in history, making them the richest and most powerful human beings in history, the average American can barely afford a tank of gas and often resorts to buying groceries on credit to feed their families.

They watch their leaders make statements instead of answering legitimate questions or taking action.

For example, if you're the CEO of a frontier AI company and you believe the technology you're building has more than a 10 percent chance of wiping out humanity, why don't you stop developing it? Once you stop, why don't you put all your vast resources into combating the threat this technology poses by building tools to counter the companies and countries that do continue to develop it?

If you're the government, why don't you raid these companies and shut them down for developing technology that might wipe out humanity? You don't need to pass a new law to do that. Why don't you ban the technology completely? If the federal government doesn't do it, why don't governors do it? Plenty of states have banned gambling, marijuana, and even pornography. Those hurt individuals. If you believe it's capable, AI won't just hurt people. It will kill us all. What are you waiting for?

Since none of that is happening, Americans rightly suspect that its politicians and tech leaders don't really believe what they're saying. If they believed it, they'd act.

All of this leads Americans to wonder: Is this an AI crisis or a leadership crisis? Meaning, is the problem the technology or the leaders behind the technology? Is there a difference between what these leaders say and what they do?

No story gets to the heart of these questions better than what happened earlier this year. On February 28, the United States conducted military strikes that resulted in the obliteration of an Iranian elementary school, killing more than 150 people, mostly children.

While the United States has not officially or publicly taken responsibility, Bloomberg reports that the Pentagon launched an investigation that, among other things, found, "Some personnel relied too much on the artificial intelligence embedded in Maven Smart System, which is made by Palantir Technologies."

According to multiple press reports, the artificial intelligence embedded in Maven Smart System was Anthropic's Claude. In fact, even after the Pentagon banned Anthropic for reasons unrelated to the elementary school bombing, Palantir was still using Anthropic. 

"The Department of War is planning to phase out Anthropic, Palantir CEO Alex Karp told CNBC on March 12, 2026. “Currently, it's not phased out. Our products are integrated with Anthropic, and in the future, it will probably be integrated with other large language models because of this dispute."

When asked by Bloomberg whether Claude was used in the elementary school bombing, Anthropic CEO Dario Amodei said he didn't know. 

Then he said this: 

"What you're talking about is a use case that doesn't even violate our red lines. We're worried that there will be a hundred times as much with use cases that do violate our red lines."

It's worth watching the full interview, which is more nuanced than that quote, but it's also worth pausing to reflect that using AI to help facilitate a strike like the bombing of an elementary school doesn't violate Anthropic's red lines.

Leading up to the bombing of the elementary school, The Washington Post reported, "As planning for a potential strike in Iran was underway, Maven, powered by Claude, suggested hundreds of targets, issued precise location coordinates, and prioritized those targets according to importance."

In a recent interview with CBS News, Amodei said, "I would rather they make fun of me than to wake up one day and discover that someone used our model, Claude, to kill a bunch of people. I don't want that."

It's entirely possible, though not officially confirmed, that Amodei has already woken up one day and discovered that his model was used to kill a bunch of people, mostly children.

Based on what Amodei has said, one might think the company would be more than happy not to be used as a weapon of war, but it's actively fighting that decision and wants the Pentagon and its partners to continue using its technology. Of course, these contracts are enormous; the last contract between Anthropic and the Pentagon was valued at up to $200M, paid for by taxpayers struggling to make ends meet.

The public is baffled by the incoherence, lack of transparency about motives, and the day-to-day whiplash between doom and utopia. To say the least. Having already written off politicians, they have simple questions for the leaders of these companies: What do you stand for? What do you actually believe? What is actually going on inside your company? What are you building exactly?

Put plainly, Americans are searching for a leader to meet this moment. Whether AI is a crisis on its own is an open question. For now, what's not is that we're in a leadership crisis, hoping someone or some entity can answer simple questions from a bewildered and beleaguered public running out of patience.

00:00:00 The Threat of Artificial Intelligence
00:01:18 Public Trust and Political Responses
00:02:03 Wealth Disparity and Leadership Inaction
00:03:49 Technology Crisis or Leadership Crisis
00:04:13 AI in Military Operations
00:05:30 Red Lines and Ethical Dilemmas
00:07:29 The Search for True Leadership

WEBVTT

00:00:00.160 --> 00:00:08.880
What<00:00:00.400> should<00:00:00.560> we<00:00:00.800> make<00:00:00.960> of<00:00:01.120> the<00:00:01.360> claim<00:00:01.760> that<00:00:02.000> AI<00:00:02.560> has<00:00:02.800> a<00:00:03.040> greater<00:00:03.280> than<00:00:03.600> 10%<00:00:04.240> chance<00:00:05.120> of<00:00:05.440> killing<00:00:05.839> all<00:00:06.160> humans<00:00:06.960> within<00:00:07.359> the<00:00:07.520> next<00:00:07.759> decade?

00:00:08.880 --> 00:00:18.000
10%<00:00:09.599> is<00:00:10.080> no<00:00:10.320> joke.<00:00:11.360> For<00:00:11.519> example,<00:00:12.000> two<00:00:12.160> weeks<00:00:12.400> before<00:00:12.639> the<00:00:12.880> 2016<00:00:13.519> election,<00:00:14.400> the<00:00:14.719> New<00:00:14.960> York<00:00:15.280> Times<00:00:16.080> gave<00:00:16.400> Donald<00:00:16.800> Trump<00:00:17.680> just

00:00:18.000 --> 00:00:21.840
a<00:00:18.400> 7%<00:00:19.760> chance<00:00:20.160> of<00:00:20.400> winning.

00:00:21.840 --> 00:00:26.720
It's<00:00:22.160> worth<00:00:23.039> repeating.<00:00:24.240> 10%<00:00:25.279> is<00:00:25.600> no<00:00:25.920> joke.

00:00:26.720 --> 00:00:30.240
Greater<00:00:27.119> than<00:00:27.359> 10%<00:00:28.080> is<00:00:28.400> greater<00:00:28.720> than<00:00:28.960> no<00:00:29.199> joke.

00:00:30.240 --> 00:00:33.680
This<00:00:30.400> isn't<00:00:30.720> coming<00:00:30.960> from<00:00:31.199> some<00:00:31.439> random<00:00:31.920> user<00:00:32.559> on<00:00:32.800> X.

00:00:33.680 --> 00:00:42.320
It's<00:00:34.000> coming<00:00:34.239> from<00:00:34.480> a<00:00:34.800> senior<00:00:35.440> safety<00:00:36.000> researcher<00:00:36.960> at<00:00:37.200> Anthropic,<00:00:38.480> the<00:00:38.800> leader<00:00:39.200> of<00:00:39.520> the<00:00:39.680> leading<00:00:40.160> frontier<00:00:40.640> AI<00:00:41.040> companies.

00:00:42.320 --> 00:00:49.280
So,<00:00:42.879> of<00:00:43.120> course,<00:00:44.160> AI<00:00:44.640> is<00:00:44.800> wildly<00:00:45.280> unpopular<00:00:45.840> with<00:00:46.079> the<00:00:46.239> very<00:00:46.559> humans<00:00:46.960> whose<00:00:47.280> lives<00:00:47.600> it<00:00:47.840> might<00:00:48.079> incinerate.

00:00:49.280 --> 00:00:58.239
According<00:00:50.160> to<00:00:50.239> a<00:00:50.480> recent<00:00:50.800> Gallup<00:00:51.200> poll,<00:00:51.920> only<00:00:52.320> 27%<00:00:53.280> of<00:00:53.520> Americans<00:00:54.480> trust<00:00:54.879> businesses<00:00:55.440> to<00:00:55.760> use<00:00:56.160> AI<00:00:56.800> responsibly.

00:00:58.239 --> 00:01:06.320
Since<00:00:58.480> that<00:00:58.719> poll<00:00:59.039> was<00:00:59.280> published,<00:01:00.079> given<00:01:00.399> the<00:01:00.559> growing<00:01:00.879> number<00:01:01.120> of<00:01:01.359> reported<00:01:01.760> incidents<00:01:02.239> inside<00:01:02.719> these<00:01:03.039> frontier<00:01:03.520> AI<00:01:04.000> companies,<00:01:04.879> what<00:01:05.040> are<00:01:05.199> the<00:01:05.360> odds<00:01:05.680> that<00:01:05.920> public

00:01:06.320 --> 00:01:14.640
trust<00:01:07.200> has<00:01:07.520> improved?<00:01:08.159> In<00:01:08.400> response,<00:01:08.960> politicians<00:01:09.600> have<00:01:09.760> released<00:01:10.159> statements,<00:01:11.200> demanded<00:01:11.760> hearings,<00:01:12.560> promoted<00:01:12.960> more<00:01:13.200> regulation,<00:01:14.320> and

00:01:14.640 --> 00:01:18.720
even<00:01:15.040> called<00:01:15.360> for<00:01:15.680> stopping<00:01:16.240> AI<00:01:16.799> development<00:01:17.360> altogether.

00:01:18.720 --> 00:01:22.960
Of<00:01:19.280> course,<00:01:19.759> this<00:01:20.080> does<00:01:20.479> nothing<00:01:20.799> to<00:01:21.040> ease<00:01:21.360> anyone's<00:01:21.920> concerns.

00:01:22.960 --> 00:01:27.200
Congressional<00:01:23.600> approval<00:01:24.080> is<00:01:24.320> lower<00:01:24.560> than<00:01:24.799> AI<00:01:25.600> at<00:01:25.920> 10%.

00:01:27.200 --> 00:01:30.560
Which<00:01:27.360> feels<00:01:27.759> 10%<00:01:28.240> too<00:01:28.560> high,<00:01:28.960> honestly.

00:01:30.560 --> 00:01:37.360
While<00:01:30.799> President<00:01:31.200> Trump<00:01:31.520> is<00:01:31.759> more<00:01:32.000> popular<00:01:32.320> than<00:01:32.640> AI<00:01:33.119> or<00:01:33.439> Congress,<00:01:34.479> he<00:01:34.799> does<00:01:34.960> not<00:01:35.200> beat<00:01:35.520> them<00:01:35.759> by<00:01:36.079> much.

00:01:37.360 --> 00:01:42.079
He's<00:01:37.759> currently<00:01:38.159> sitting<00:01:38.400> at<00:01:38.640> a<00:01:38.960> 32%<00:01:40.079> approval<00:01:40.560> rating.

00:01:42.079 --> 00:01:51.119
This<00:01:42.320> week,<00:01:42.560> he<00:01:42.799> got<00:01:42.960> the<00:01:43.119> leaders<00:01:43.439> of<00:01:43.520> these<00:01:43.759> companies<00:01:44.159> to<00:01:44.320> come<00:01:44.479> to<00:01:44.560> the<00:01:44.720> White<00:01:44.960> House,<00:01:45.840> sign<00:01:46.159> a<00:01:46.399> document,<00:01:47.360> and<00:01:47.680> stand<00:01:48.320> awkwardly<00:01:49.439> in<00:01:49.680> front<00:01:49.759> of<00:01:49.840> the<00:01:50.079> press.

00:01:51.119 --> 00:01:59.360
He<00:01:51.439> also<00:01:51.920> rebranded<00:01:52.560> AI<00:01:53.119> as<00:01:53.520> super<00:01:53.920> intelligence,<00:01:55.119> hoping<00:01:55.520> America's<00:01:56.079> love<00:01:56.320> for<00:01:56.720> superhero<00:01:57.439> movies<00:01:58.479> will<00:01:58.799> change<00:01:59.119> how

00:01:59.360 --> 00:02:04.799
it<00:01:59.600> feels<00:01:59.840> about<00:02:00.159> things.<00:02:01.840> It's<00:02:02.159> worth<00:02:02.479> noting<00:02:02.960> where<00:02:03.280> we<00:02:03.520> are<00:02:03.759> right<00:02:04.000> now.

00:02:04.799 --> 00:02:11.840
While<00:02:04.960> the<00:02:05.119> purveyors<00:02:05.600> of<00:02:05.759> this<00:02:05.920> technology<00:02:06.479> build<00:02:06.719> the<00:02:06.960> most<00:02:07.119> valuable<00:02:07.520> companies<00:02:07.840> in<00:02:08.080> history,<00:02:08.879> making<00:02:09.119> them<00:02:09.280> the<00:02:09.520> richest<00:02:09.840> and<00:02:10.080> most<00:02:10.319> powerful<00:02:10.879> human<00:02:11.280> beings<00:02:11.599> in

00:02:11.840 --> 00:02:20.480
history,<00:02:12.879> the<00:02:13.200> average<00:02:13.520> American<00:02:14.080> can<00:02:14.319> barely<00:02:14.640> afford<00:02:15.040> a<00:02:15.280> tank<00:02:15.520> of<00:02:15.680> gas<00:02:16.720> and<00:02:16.959> often<00:02:17.360> resorts<00:02:17.760> to<00:02:18.000> buying<00:02:18.400> groceries<00:02:18.959> on<00:02:19.280> credit<00:02:20.160> to

00:02:20.480 --> 00:02:30.160
feed<00:02:20.720> their<00:02:20.959> families.<00:02:22.239> They<00:02:22.560> watch<00:02:22.879> their<00:02:23.120> leaders<00:02:23.599> make<00:02:23.920> statements<00:02:25.120> instead<00:02:25.440> of<00:02:25.680> answering<00:02:26.160> legitimate<00:02:26.800> questions<00:02:27.840> or<00:02:28.160> taking<00:02:28.560> action.

00:02:30.160 --> 00:02:38.480
For<00:02:30.400> example,<00:02:31.520> if<00:02:31.760> you're<00:02:32.000> the<00:02:32.239> CEO<00:02:32.800> of<00:02:33.040> a<00:02:33.360> frontier<00:02:33.840> AI<00:02:34.319> company<00:02:35.120> and<00:02:35.440> you<00:02:35.680> believe<00:02:35.920> the<00:02:36.160> technology<00:02:36.640> you're<00:02:36.959> building<00:02:37.599> has<00:02:37.920> more<00:02:38.080> than<00:02:38.239> a

00:02:38.480 --> 00:02:45.599
10%<00:02:39.200> chance<00:02:40.000> of<00:02:40.400> wiping<00:02:40.879> out<00:02:41.200> humanity,<00:02:42.879> why<00:02:43.040> don't<00:02:43.280> you<00:02:43.440> stop<00:02:43.760> developing<00:02:44.319> it?

00:02:45.599 --> 00:02:54.480
Once<00:02:45.920> you<00:02:46.080> stop,<00:02:46.720> why<00:02:46.959> don't<00:02:47.120> you<00:02:47.280> put<00:02:47.599> all<00:02:47.840> your<00:02:48.080> vast<00:02:48.560> resources<00:02:49.680> into<00:02:50.080> combating<00:02:50.480> the<00:02:50.720> threat<00:02:51.040> this<00:02:51.280> technology<00:02:51.920> poses<00:02:52.720> by<00:02:53.040> building<00:02:53.440> tools<00:02:54.239> to

00:02:54.480 --> 00:03:00.640
counter<00:02:54.800> the<00:02:55.040> companies<00:02:55.360> and<00:02:55.680> countries<00:02:56.640> that<00:02:57.040> do<00:02:57.920> continue<00:02:58.560> to<00:02:58.879> develop<00:02:59.280> it?

00:03:00.640 --> 00:03:07.200
If<00:03:00.800> you're<00:03:01.040> the<00:03:01.200> government,<00:03:01.920> why<00:03:02.080> don't<00:03:02.239> you<00:03:02.400> raid<00:03:02.640> these<00:03:02.959> companies<00:03:03.200> and<00:03:03.440> shut<00:03:03.680> them<00:03:03.840> down<00:03:04.000> for<00:03:04.239> developing<00:03:04.720> technology<00:03:05.200> that<00:03:05.440> might<00:03:05.680> wipe<00:03:05.920> out<00:03:06.159> humanity?

00:03:07.200 --> 00:03:09.280
You<00:03:07.360> don't<00:03:07.519> need<00:03:07.599> to<00:03:07.760> pass<00:03:07.920> a<00:03:08.080> new<00:03:08.239> law<00:03:08.400> to<00:03:08.640> do<00:03:08.720> that.

00:03:09.280 --> 00:03:11.920
Why<00:03:09.440> don't<00:03:09.599> you<00:03:09.760> ban<00:03:10.000> the<00:03:10.239> technology<00:03:10.879> completely?

00:03:11.920 --> 00:03:16.000
If<00:03:12.159> the<00:03:12.319> federal<00:03:12.720> government<00:03:13.040> doesn't<00:03:13.360> do<00:03:13.519> it,<00:03:14.080> why<00:03:14.319> don't<00:03:14.640> governors<00:03:15.200> do<00:03:15.360> it?

00:03:16.000 --> 00:03:22.159
Plenty<00:03:16.319> of<00:03:16.480> states<00:03:16.800> have<00:03:17.040> banned<00:03:17.360> gambling,<00:03:18.480> marijuana,<00:03:19.599> and<00:03:19.920> even<00:03:20.480> pornography.

00:03:22.159 --> 00:03:27.519
Those<00:03:22.560> hurt<00:03:23.200> individuals<00:03:24.560> if<00:03:24.959> you<00:03:25.200> believe<00:03:25.440> it's<00:03:25.840> capable.

00:03:27.519 --> 00:03:32.239
AI<00:03:28.000> won't<00:03:28.319> just<00:03:28.640> hurt<00:03:28.959> people,<00:03:29.760> it<00:03:30.000> will<00:03:30.239> kill<00:03:30.480> us<00:03:30.799> all.

00:03:32.239 --> 00:03:40.720
What<00:03:32.400> are<00:03:32.560> you<00:03:32.799> waiting<00:03:33.120> for?<00:03:34.879> Since<00:03:35.200> none<00:03:35.440> of<00:03:35.519> that<00:03:35.680> is<00:03:35.920> happening,<00:03:37.120> Americans<00:03:37.760> rightly<00:03:38.319> suspect<00:03:38.959> that<00:03:39.280> it's<00:03:39.599> politicians<00:03:40.159> and<00:03:40.400> tech

00:03:40.720 --> 00:03:45.680
leaders<00:03:42.080> don't<00:03:42.480> really<00:03:42.959> believe<00:03:43.840> what<00:03:44.159> they're<00:03:44.480> saying.

00:03:45.680 --> 00:03:49.519
If<00:03:45.920> they<00:03:46.159> believed<00:03:46.480> it,<00:03:47.440> they'd<00:03:47.840> act.

00:03:49.519 --> 00:03:56.159
All<00:03:49.760> of<00:03:49.920> this<00:03:50.239> leads<00:03:50.560> Americans<00:03:51.040> to<00:03:51.280> wonder,<00:03:52.080> is<00:03:52.239> this<00:03:52.480> an<00:03:52.720> AI<00:03:53.200> crisis<00:03:54.080> or<00:03:54.400> a<00:03:54.640> leadership<00:03:55.200> crisis?

00:03:56.159 --> 00:04:02.560
Meaning,<00:03:56.959> is<00:03:57.120> the<00:03:57.599> problem<00:03:58.400> the<00:03:58.720> technology<00:03:59.760> or<00:04:00.080> the<00:04:00.319> leaders<00:04:00.879> behind<00:04:01.360> the<00:04:01.599> technology?

00:04:02.560 --> 00:04:07.519
Is<00:04:02.799> there<00:04:02.959> a<00:04:03.200> difference<00:04:03.680> between<00:04:04.000> what<00:04:04.319> these<00:04:04.640> leaders<00:04:05.120> say<00:04:06.080> and<00:04:06.400> what<00:04:06.640> they<00:04:06.879> do?

00:04:07.519 --> 00:04:13.760
No<00:04:07.920> story<00:04:08.640> gets<00:04:08.959> to<00:04:09.120> the<00:04:09.280> heart<00:04:09.439> of<00:04:09.599> these<00:04:09.840> questions<00:04:10.239> better<00:04:11.040> than<00:04:11.280> what<00:04:11.599> happened<00:04:12.000> earlier<00:04:12.480> this<00:04:12.720> year.

00:04:13.760 --> 00:04:22.720
On<00:04:14.000> February<00:04:14.560> 28th,<00:04:15.120> the<00:04:15.360> United<00:04:15.760> States<00:04:16.160> conducted<00:04:16.799> military<00:04:17.280> strikes<00:04:17.680> that<00:04:17.919> resulted<00:04:18.400> in<00:04:18.560> the<00:04:18.720> obliteration<00:04:19.840> of<00:04:20.079> an<00:04:20.239> Iranian<00:04:21.040> elementary<00:04:21.600> school,<00:04:22.320> killing

00:04:22.720 --> 00:04:27.600
more<00:04:22.960> than<00:04:24.160> 150<00:04:24.560> people,<00:04:25.600> mostly<00:04:26.960> children.

00:04:27.600 --> 00:04:36.400
While<00:04:27.840> the<00:04:28.160> United<00:04:28.479> States<00:04:28.800> has<00:04:29.040> not<00:04:29.199> officially<00:04:29.759> or<00:04:30.080> publicly<00:04:30.639> taken<00:04:31.040> responsibility,<00:04:32.240> Bloomberg<00:04:32.880> reports<00:04:33.280> that<00:04:33.440> the<00:04:33.680> Pentagon<00:04:34.560> launched<00:04:34.960> an<00:04:35.199> investigation<00:04:35.919> that,

00:04:36.400 --> 00:04:45.440
among<00:04:36.720> other<00:04:37.040> things,<00:04:37.440> found<00:04:38.479> some<00:04:38.800> personnel<00:04:39.440> relied<00:04:40.000> too<00:04:40.240> much<00:04:40.560> on<00:04:40.800> the<00:04:40.960> artificial<00:04:41.600> intelligence<00:04:42.639> embedded<00:04:43.280> in<00:04:43.600> Maven<00:04:44.240> Smart<00:04:44.560> System,

00:04:45.440 --> 00:04:48.639
which<00:04:45.680> is<00:04:45.919> made<00:04:46.639> by<00:04:46.960> Palantir<00:04:47.759> Technologies.

00:04:48.639 --> 00:04:57.360
According<00:04:49.280> to<00:04:49.520> multiple<00:04:50.000> press<00:04:50.320> reports,<00:04:51.199> the<00:04:51.440> artificial<00:04:52.080> intelligence<00:04:52.800> embedded<00:04:53.199> in<00:04:53.360> Maven<00:04:53.919> Smart<00:04:54.240> System<00:04:54.800> was<00:04:55.199> Anthropic's<00:04:56.080> Claude.

00:04:57.360 --> 00:05:06.160
In<00:04:57.600> fact,<00:04:58.000> even<00:04:58.320> after<00:04:58.560> the<00:04:58.800> Pentagon<00:04:59.280> banned<00:04:59.680> Anthropic<00:05:00.400> for<00:05:00.720> reasons<00:05:01.120> unrelated<00:05:01.680> to<00:05:01.840> the<00:05:02.080> elementary<00:05:02.479> school<00:05:02.800> bombing,<00:05:03.840> Palantir<00:05:04.720> was<00:05:05.040> still<00:05:05.759> using

00:05:06.160 --> 00:05:14.479
Anthropic.<00:05:07.600> Here's<00:05:07.919> what<00:05:08.160> Palantir<00:05:08.800> CEO<00:05:09.280> Alex<00:05:09.680> Karp<00:05:10.400> told<00:05:10.639> CNBC<00:05:11.919> on<00:05:12.240> March<00:05:12.560> 12th,<00:05:13.360> 2026.

00:05:14.479 --> 00:05:17.440
<00:05:14.695> The<00:05:14.911> Department<00:05:15.127> of<00:05:15.343> War<00:05:15.559> is<00:05:15.775> planning<00:05:15.991> to<00:05:16.207> phase<00:05:16.423> out<00:05:16.639> Anthropic.

00:05:17.440 --> 00:05:25.759
Currently,<00:05:17.840> it's<00:05:18.160> not<00:05:18.240> phased<00:05:18.639> out.<00:05:19.039> Our<00:05:19.360> products<00:05:19.759> are<00:05:20.000> integrated<00:05:20.720> with<00:05:21.039> Anthropic<00:05:22.080> and<00:05:22.320> and<00:05:23.440> uh<00:05:23.600> and<00:05:23.840> in<00:05:24.080> the<00:05:24.240> future,<00:05:24.720> it<00:05:25.039> will<00:05:25.440> probably

00:05:25.759 --> 00:05:30.160
be<00:05:26.115> integrated<00:05:26.472> with<00:05:26.828> other<00:05:27.185> large<00:05:27.541> language<00:05:27.898> models<00:05:28.254> because<00:05:28.611> of<00:05:28.967> of<00:05:29.324> this<00:05:29.680> dispute.

00:05:30.160 --> 00:05:39.360
<00:05:30.545> When<00:05:30.930> asked<00:05:31.314> by<00:05:31.699> Bloomberg<00:05:32.084> whether<00:05:32.469> Claude<00:05:32.853> was<00:05:33.238> used<00:05:33.623> in<00:05:34.008> the<00:05:34.392> elementary<00:05:34.777> school<00:05:35.162> bombing,<00:05:35.547> Anthropic<00:05:35.931> CEO<00:05:36.316> Dario<00:05:36.701> Amodei<00:05:37.086> said<00:05:37.470> he<00:05:37.855> didn't<00:05:38.240> know.

00:05:39.360 --> 00:05:45.759
Then<00:05:40.400> he<00:05:40.720> said<00:05:40.960> this,<00:05:41.520> "What<00:05:41.919> what<00:05:42.160> you're<00:05:42.479> talking<00:05:42.720> about<00:05:43.280> is<00:05:43.440> a<00:05:43.680> use<00:05:43.840> case<00:05:44.000> that<00:05:44.240> doesn't<00:05:44.479> even<00:05:44.639> violate<00:05:45.039> our<00:05:45.199> red<00:05:45.440> lines.

00:05:45.759 --> 00:05:52.639
We're<00:05:46.000> worried<00:05:46.320> that<00:05:46.479> there<00:05:46.639> will<00:05:46.880> be<00:05:47.039> a<00:05:47.120> hundred<00:05:47.440> times<00:05:47.759> as<00:05:48.080> much,<00:05:48.720> you<00:05:48.880> know,<00:05:49.360> with<00:05:49.600> with<00:05:50.000> use<00:05:50.240> cases<00:05:50.639> that<00:05:50.800> that<00:05:50.880> that<00:05:51.360> do<00:05:51.600> violate<00:05:51.919> our<00:05:52.160> red<00:05:52.320> line."

00:05:52.639 --> 00:05:56.880
<00:05:52.934> It's<00:05:53.230> worth<00:05:53.525> watching<00:05:53.821> the<00:05:54.116> full<00:05:54.411> interview,<00:05:54.707> which<00:05:55.002> is<00:05:55.297> more<00:05:55.593> nuanced<00:05:55.888> than<00:05:56.184> that<00:05:56.479> clip.

00:05:56.880 --> 00:06:05.600
And<00:05:57.039> we<00:05:57.199> link<00:05:57.440> to<00:05:57.520> the<00:05:57.680> full<00:05:57.919> interview<00:05:58.160> in<00:05:58.400> the<00:05:58.479> show<00:05:58.720> notes,<00:05:59.440> but<00:05:59.680> it's<00:06:00.000> also<00:06:00.400> worth<00:06:00.720> pausing<00:06:01.759> and<00:06:02.080> reflecting<00:06:02.479> on<00:06:02.720> the<00:06:02.880> fact<00:06:02.960> that<00:06:03.280> using<00:06:03.680> AI<00:06:04.560> to<00:06:04.800> help<00:06:05.039> facilitate

00:06:05.600 --> 00:06:12.880
a<00:06:05.840> strike<00:06:06.160> like<00:06:06.319> the<00:06:06.479> bombing<00:06:06.800> of<00:06:06.960> an<00:06:07.199> elementary<00:06:07.680> school<00:06:08.479> doesn't<00:06:09.039> violate<00:06:09.840> Anthropic's<00:06:10.639> red<00:06:10.960> lines.

00:06:12.880 --> 00:06:21.759
Leading<00:06:13.199> up<00:06:13.360> to<00:06:13.440> the<00:06:13.600> bombing<00:06:13.919> of<00:06:14.080> the<00:06:14.240> elementary<00:06:14.639> school,<00:06:14.960> the<00:06:15.199> Washington<00:06:15.600> Post<00:06:16.000> reported<00:06:16.880> as<00:06:17.199> planning<00:06:17.600> for<00:06:17.759> a<00:06:18.000> potential<00:06:18.479> strike<00:06:18.800> in<00:06:19.039> Iran<00:06:19.440> was<00:06:19.680> underway,<00:06:20.720> Maven,

00:06:21.759 --> 00:06:30.319
powered<00:06:22.160> by<00:06:22.479> Claude,<00:06:23.600> suggested<00:06:24.319> hundreds<00:06:24.720> of<00:06:24.880> targets,<00:06:25.919> issued<00:06:26.479> precise<00:06:27.199> location<00:06:27.759> coordinates,<00:06:28.960> and<00:06:29.280> prioritized<00:06:30.000> those

00:06:30.319 --> 00:06:32.800
targets<00:06:31.280> according<00:06:31.759> to<00:06:32.000> importance.

00:06:32.800 --> 00:06:37.120
In<00:06:33.039> a<00:06:33.199> recent<00:06:33.520> interview<00:06:33.840> with<00:06:34.080> CBS<00:06:34.639> News,<00:06:35.520> Amodei<00:06:36.160> said<00:06:36.319> this.

00:06:37.120 --> 00:06:44.880
<00:06:37.365> I<00:06:37.611> would<00:06:37.856> rather<00:06:38.101> they<00:06:38.347> make<00:06:38.592> fun<00:06:38.837> of<00:06:39.082> me<00:06:39.328> than<00:06:39.573> to<00:06:39.818> wake<00:06:40.064> up<00:06:40.309> one<00:06:40.554> day<00:06:40.799> and<00:06:41.045> and<00:06:41.290> discover<00:06:41.535> that<00:06:41.781> someone<00:06:42.026> used<00:06:42.271> our<00:06:42.517> our<00:06:42.762> model<00:06:43.007> Claude<00:06:43.253> to<00:06:43.498> kill<00:06:43.743> a<00:06:43.988> bunch<00:06:44.234> of<00:06:44.479> people.

00:06:44.880 --> 00:06:53.280
I<00:06:45.261> don't<00:06:45.642> want<00:06:46.023> that.<00:06:46.404> <00:06:46.785> It's<00:06:47.166> entirely<00:06:47.547> possible,<00:06:47.928> though<00:06:48.309> not<00:06:48.690> officially<00:06:49.070> confirmed,<00:06:49.451> that<00:06:49.832> Amodei<00:06:50.213> has<00:06:50.594> already<00:06:50.975> woken<00:06:51.356> up<00:06:51.737> one<00:06:52.118> day<00:06:52.499> and<00:06:52.880> discovered

00:06:53.280 --> 00:06:58.639
that<00:06:53.520> his<00:06:53.840> model<00:06:54.479> was<00:06:54.800> used<00:06:55.039> to<00:06:55.280> kill<00:06:55.440> a<00:06:55.680> bunch<00:06:55.840> of<00:06:56.000> people,<00:06:56.960> mostly<00:06:57.440> children.

00:06:58.639 --> 00:07:07.199
Based<00:06:58.960> on<00:06:59.120> what<00:06:59.360> Amodei<00:06:59.840> has<00:07:00.080> said,<00:07:00.400> one<00:07:00.639> might<00:07:00.880> think<00:07:01.039> the<00:07:01.280> company<00:07:01.599> would<00:07:01.840> be<00:07:02.000> more<00:07:02.240> than<00:07:02.479> happy<00:07:03.280> not<00:07:03.520> to<00:07:03.680> be<00:07:03.840> used<00:07:04.000> as<00:07:04.240> a<00:07:04.319> weapon<00:07:04.560> of<00:07:04.800> war,<00:07:05.759> but<00:07:06.000> it's<00:07:06.240> actively<00:07:06.880> fighting

00:07:07.199 --> 00:07:14.400
that<00:07:07.520> decision<00:07:08.400> and<00:07:08.720> wants<00:07:08.960> the<00:07:09.199> Pentagon<00:07:09.759> and<00:07:10.000> its<00:07:10.240> partners<00:07:10.720> to<00:07:11.039> continue<00:07:12.080> using<00:07:12.400> its<00:07:12.720> technology.

00:07:14.400 --> 00:07:17.120
Of<00:07:14.639> course,<00:07:14.960> these<00:07:15.280> contracts<00:07:15.840> are<00:07:16.080> enormous.

00:07:17.120 --> 00:07:26.240
The<00:07:17.360> last<00:07:17.680> contract<00:07:18.160> between<00:07:18.560> Anthropic<00:07:19.199> and<00:07:19.360> the<00:07:19.440> Pentagon<00:07:20.000> was<00:07:20.240> valued<00:07:20.560> at<00:07:20.800> up<00:07:20.960> to<00:07:21.280> $200<00:07:21.919> million,<00:07:23.199> paid<00:07:23.520> for<00:07:23.759> by<00:07:24.080> taxpayers<00:07:25.280> struggling

00:07:26.240 --> 00:07:34.639
to<00:07:26.479> make<00:07:26.720> ends<00:07:27.039> meet.<00:07:28.720> The<00:07:29.039> public<00:07:29.919> is<00:07:30.240> baffled<00:07:31.120> by<00:07:31.360> the<00:07:31.599> incoherence,<00:07:33.199> lack<00:07:33.440> of<00:07:33.680> transparency<00:07:34.319> about

00:07:34.639 --> 00:07:40.639
motives,<00:07:35.759> and<00:07:36.000> the<00:07:36.160> day-to-day<00:07:36.720> whiplash<00:07:37.520> between<00:07:37.919> doom<00:07:38.960> and<00:07:39.360> utopia.

00:07:40.639 --> 00:07:48.639
To<00:07:40.880> say<00:07:41.039> the<00:07:41.199> least,<00:07:41.919> having<00:07:42.319> already<00:07:42.639> written<00:07:42.960> off<00:07:43.199> politicians,<00:07:43.840> they<00:07:44.080> have<00:07:44.400> simple<00:07:44.960> questions<00:07:46.000> for<00:07:46.319> the<00:07:46.479> leaders<00:07:46.800> of<00:07:46.960> these<00:07:47.280> companies.

00:07:48.639 --> 00:07:52.080
What<00:07:48.800> do<00:07:48.960> you<00:07:49.199> stand<00:07:49.360> for?<00:07:50.240> What<00:07:50.400> do<00:07:50.479> you<00:07:50.639> actually<00:07:50.960> believe?

00:07:52.080 --> 00:07:55.759
What<00:07:52.319> is<00:07:52.560> actually<00:07:52.960> going<00:07:53.199> on<00:07:53.440> inside<00:07:53.840> your<00:07:54.240> company?

00:07:55.759 --> 00:07:59.440
What<00:07:56.000> are<00:07:56.160> you<00:07:56.479> building<00:07:57.759> exactly?

00:07:59.440 --> 00:08:05.680
Put<00:07:59.759> plainly,<00:08:01.120> Americans<00:08:01.680> are<00:08:01.919> searching<00:08:02.240> for<00:08:02.479> a<00:08:02.720> leader<00:08:03.360> to<00:08:03.680> meet<00:08:04.000> this<00:08:04.319> moment.

00:08:05.680 --> 00:08:11.199
Whether<00:08:06.080> AI<00:08:06.479> is<00:08:06.639> a<00:08:06.879> crisis<00:08:07.199> on<00:08:07.360> its<00:08:07.680> own<00:08:08.319> is<00:08:08.560> an<00:08:08.800> open<00:08:09.120> question<00:08:10.400> for<00:08:10.720> now.

00:08:11.199 --> 00:08:19.520
What's<00:08:11.520> not<00:08:12.000> is<00:08:12.240> that<00:08:12.479> we're<00:08:12.720> in<00:08:12.960> a<00:08:13.199> leadership<00:08:13.840> crisis,<00:08:14.879> hoping<00:08:15.360> someone<00:08:15.840> or<00:08:16.240> some<00:08:16.479> entity<00:08:16.960> can<00:08:17.280> answer<00:08:17.680> simple<00:08:18.160> questions<00:08:19.120> from<00:08:19.360> a

00:08:19.520 --> 00:08:25.520
bewildered<00:08:20.639> and<00:08:20.879> beleaguered<00:08:21.599> public<00:08:22.879> running<00:08:23.199> out<00:08:23.360> of<00:08:23.520> patience.