Skip to content
TrackPodcasts
newsApr 22, 202622:49

White hat, black box: AI’s next chapter

About this episode

The decision of Anthropic, an AI giant, to keep its Mythos model sequestered surely makes for good press. But there seems to be more to it than that—and it might change the whole industry’s approach. Indian politicians are chasing female voters more than ever; we question the means and the outcomes. And next in our World Cup contender-country profiles: Senegal. 


Guests and host:

  • Alex Hern, AI writer
  • Kira Huju, Asia correspondent
  • Jon Fasman, senior culture correspondent
  • Jason Palmer, co-host of “The Intelligence”


Topics covered: 

  • AI, Anthropic, Mythos
  • India, women, politics
  • World Cup, Senegal


Get a world of insights by subscribing to Economist Podcasts+. For more information about how to access Economist Podcasts+, please visit our FAQs page or watch our video explaining how to link your account.



Hosted on Acast. See acast.com/privacy for more information.

Interactive timestamps

Jump to segment

Get every episode summarized

Each time The Intelligence from The Economist publishes, we email you a written briefing from the transcript — the topics, who appeared, and any specific claims, with the ad reads skipped.

Email me new episodes

Free for 3 shows. No card needed.

Hosts & guests

Transcript ready

208 searchable segments. Every word is indexed and playable.

White hat, black box: AI’s next chapter

The Intelligence from The Economist

0:00
22:49

Full transcript

The Intelligence from The EconomistWhite hat, black box: AI’s next chapter. Machine-transcribed; use the interactive transcript above to jump the player to any line.

0:00Hello and welcome to the Intelligence from the Economist. I'm your host, Jason Palmer. Every weekday we provide a fresh perspective on the events shaping your world. In India, there's a big political shift underway. The country's women are becoming an ever-more potent force in the electorate. Whether what politicians are doing to win them over is vote-buying or genuinely empowering is not entirely clear. And my former co-host and a sports nut, John Fasman, is leading us up to the World Cup by profiling the countries of 10 teams that are in contention. This week, it's Senegal. First up, though.

1:11Normally, when someone tells me they've got a new, superhuman AI, but I'm not allowed to see it, they're telling tall tales. But with Anthropics' new Mythos AI system, there's a chance they're telling the truth. Alex Hurn is the economist, AI writer. It's a superhuman hacker that's so dangerous they can't release it. It's forcing skeptics to accept that some of the dangers of AI may come sooner than they expected. And it's also making the AI industry reassess its own business model. So let's start with this new piece of mind-bending tech. What is Mythos? Club Mythos is the latest AI system from Anthropic, the makers of Claude. It's, as the name perhaps suggests, bigger and better than its previous models, Haikus, Sonnet and Opus. Those are getting larger each time, and Mythos is the biggest yet. But unlike those previous models, it's not been released. Anthropic went to all the effort of training it, and then as a preview version only, gave it to a select group of partners,

2:1511 named groups and 40 smaller organizations. And that's because it is, Anthropic says, a tremendously good hacker. You pour money in one end, you get software vulnerabilities out the other. And that means Anthropic thinks it's too dangerous to make available for general release. So are you one of the mysterious 40 odd entities who has received the model? Have you had a play? I am not. I have previously been given access to models that Dary Amadei, now the head of Anthropic, says are too dangerous for public release. But this time, no, it really is kept very much behind closed doors. But we have seen evidence of the claims that Anthropic are making. They have been putting it to work, finding software vulnerabilities in open source code, things like operating systems and media encoders. And they've published a couple of the ones they found already, a complex bug in the operating system open BSD that hadn't unchanged for 27 years until Mythos discovered it. And so essentially the rest of those companies are being given access

3:15so they can find the holes in their own stuff and patch them before a general release. Is that the idea? Yeah, exactly. So Anthropic picked 11 companies, including Apple, Microsoft, NVIDIA, JP Morgan, the only financial services provider, Palo Alto Networks, Broadcom, there's 11 of them. And it tried to select companies with systemic importance, operating system manufacturers, browser manufacturers, big financial institutions. And the idea is partially to check that Mythos can do what it can do. And also just to give these huge systemic institutions time to build their defenses up before Mythos or something of equivalent quality is made available to people who might use it as an attacker. And you're putting aside your skepticism for the moment. I mean, it has to be said that saying it's this badass suggests that we have the best thing in the world and you should pay for it. Absolutely. And there are a lot of reasons to be skeptical. This plays into Anthropics entire bit very well. They are not only keen to present themselves as

4:16the makers of the best coding software in the world. And that's all hacking really is right, a specialized form of coding. They also like presenting themselves as the most safety-oriented lab. And this is the most safety-oriented move you can do. There's also lesser advantages. It lets them handle a compute crunch that they're going through with elegance and grace. This is a big expensive to run model. It seems likely that we're anthropic to make it available for general release. They don't have the GPUs to run it at the level that their customers would want. They will be forced to put prices up in a very obvious rationing move or just ration it directly with usage limits. So this sort of stage release lets them show off that they've got all of this power without admitting that they don't have the power to run it. And then also it prevents the photocopies from being started up. If you make it available to a select group of partners, those partners are trustworthy organizations who aren't going to take your intellectual property, feed it in as training data to their own models and produce a fast follow-in three to six months, which is exactly what Anthropic has already accused some of the

5:20largest Chinese labs of doing with its previous opus models. I suppose though this brings us into an era that lots of people had said was coming. The idea of the dual use nature of AI coming to bite us at some point when it got good enough. Exactly. And this is one of the reasons why I'm less skeptical than that long list of reasons why this is the sort of thing Anthropic would love to do might suggest. We've known this is coming. AI systems have been getting better and better at software engineering for the last three years. They've passed a load of qualitative thresholds already and it was a matter of time until a system was created that was a very capable hacker because that's what happens when you get better and better at software engineering. There were only really a few open questions. One was whether or not you could make a system that was a good hacker and teach it the ethics required to not be a malicious hacker. And the other was if someone made a system that was a good hacker and couldn't teach it there's ethics, whether they would release it

6:21and Anthropic shows that one possible answer here is they cannot yet teach Mythos to be safe but they can keep it behind closed doors. So the next big question is whether we get these fast followers eventually catching up and if anyone cracks the how to make this a safe thing to release before someone gives in and just releases the unsafe version to the public at large. But in a number of ways this doesn't seem very sustainable or very equitable I suppose. Well, the companies that are inside the tent must be very pleased to get this early access but it can't stay that way forever. Right. I mean whether or not it can stay that way forever is I think still an open question. It's not clear right now how quickly other people will catch up to Anthropic and at least in the US there seems to be broad agreement that general release for a model of this class isn't a smart thing to do. Open AI is making hay about the fact that it wants a more democratic release schedule but even that democratic release schedule still involves verifying people's real identities and checking they work inside of security and so on. The question hovering over everyone's head is whether or not the Chinese labs will follow.

7:26They tend to have had a more laissez faire approach to safety but at the same time the Chinese government is making noises of concern. Everyone wants this sort of power but I don't think anyone wants to be responsible for unleashing an era of all systems being hackable by all people. So there's a chance that some form of cautious release will become the norm in this space. But then yes the power dynamic that creates ought to I think concern a lot of us. If you need to prove your credentials to access a system that is required to secure your own networks then that's inherently an advantage of boost to those who can provide their credentials. It harms startups, it harms people getting their leg up in the industry and it helps people who are ready on top. There are to ask how regulation might come into this to even the score to keep everybody honest. Well I mean the good news is this has provided a pretty obvious model for any government body to copy. Until anthropic voluntarily did this behind closed doors access it would

8:29have been a big thing for a government agency to enforce. Now that they've done it it's somewhat easier for a regulator to look at the other labs and go follow their lead or will come down on you. A lot of government regulation begins as voluntary action from a sector and then it's rarefied over time and I think this will be no different. The question yes is whether or not the government feels like it has the responsibility or authority to verify who those insiders are. It adds democratic legitimacy if it does but also has the uncomfortable overtones of a government picking winners choosing insiders and banishing outsiders. One reason why they may well want to do that is the whole thing has national security implications. You obviously want your software and that of your allies to be better protected than those of the people you are trying to hack at the same time. The American intelligence community has always struggled with this dual mandate that it protects American cyber security while also hoarding the ability to use American technology to hack its foes and I think this is no different. Alex thanks very much for your time. Thank you for

9:33having me. A couple of weeks ago I attended an election rally on the outskirts of Kolkata in the state of West Bengal and asked the rally goers for the ruling Trinimal Congress assembled on this narrow

10:33dirt road. It occurred to me that what I was looking at was a perfect visualization of the trend that I'd come here to witness which was that liking in the elections women had moved center stage. At the front you had an all-female marching block and they were dressed in bold prints, they were loudly chanting their way through the town and then the men had gathered at the back almost like the women's shadow. Now this has been a busy month for state elections in India. We have four states and one union territory voting across the country and in West Bengal the election is taking place in two stages the first one happening tomorrow and then the second one in April the 29th and what all of these elections turn out to have in common is a striking demographic shift. Women are determining and changing how Indian elections are fought.

11:38So that prominence of women in the electoral sphere is a new thing for India. It's most definitely new thing in India's first general election in the early 50s you actually had a situation where three million Indian women were struck off the voter list because it had not occurred to them to register themselves with their own names they had simply registered as wives or daughters of male family members and now you have a situation where for the first time in 2019 female turnout actually surpassed male turnout and that's something that we saw again in the 2024 national election and so Indian women have gone from being a minor figure to being entirely central to who wins elections. And so this is just a straightforward story of modern women's empowerment in India. It is definitely the case that Indian women tend to vote more when they have a job when they get a better education when they're more politically exposed and so in many ways yes voting reflects women's improved plays in society but I can say having

12:40gone through the sweat of precession in West Bengal that it was not in fact all of feminist post grads. I spoke to a march over who was a jobless widow with a disabled daughter. I spoke to a Dalit woman formerly known as an outcast who was struggling to and produce the right paperwork to vote in the upcoming election. Also what's happening in these kinds of spaces also suggest to you that there is a way in which Indian politicians have figured out how to speak to marginalized women as well. What do you mean that politicians have learned how to do that? What are they doing? We have some really great research by Sanjay Kumar and colleagues at the Centre for the Study of Developing Societies which is India's Premier Social Science Institute and that research has suggested that women are making Indian elections more pragmatic women are much more driven by really concrete tangible welfare policies less than 10% are interested in voting based on ideology or culture war issues and that's quite distinct from the way that Indian men often vote.

13:42And so election strategy sat down and said all right what is the most tangible thing that you can possibly do in terms of welfare for women and where we seem to have ended up is cash sending women monthly cash transfers. All of those programs look like in practice on the ground. So the sums that are supplied range from about 800 to 2500 rupees per month that's about 9 to 27 dollars which may not sound like a lot but in India it is quite significant given that Indian women are half as likely to hold a job as men and so these kinds of sums are in fact quite life changing. So what you have is now at least 16 out of India's 28 states have female only direct cash transfer schemes and that's up from just a handful in 2022 and suspiciously many of these were introduced just before election day. Are they means tested? So they are never

14:44means tested. There are very in criteria depending on the scheme age income marital status but once you've cleared those there is nothing you have to do nothing you have to prove that you're doing and that might just get sent to you. But what about the notion that it is or at least looks like vote buying? There is a lot of interesting social scientific research that suggests that some of these schemes can be very beneficial. If you look up what's going on even just in these state elections though a lot of it does look very have hazard so move away from West Bengal to the state of Tamil Nadu which is the other place where we have voting on the same day is now in West Bengal and here Chief Minister MK Starling in anticipation of the elections has dispatched what he calls a summer special package worth about $50 to more than 13 million women in Tamil Nadu. He's promised also to double the states all female cash grant if reelected this is a pledge that was quickly matched or out bit by all the opposition parties. Successful it might be it does sound

15:48expensive though. Yeah the cash friends is getting a little out of hand in the last financial year the states that have these schemes spent more than $18 billion worth on unconditional cash transfer programs mostly for women. If you look at proffedigate West Bengal in particular where state government debt is 38% of the state's GDP which is approaching record levels. Lakshmi Bandar alone which is the flagship cash transfer scheme consumes 10% of the state's revenue receipts and that's just one of West Bengal's many many such program. I guess the question putting the politics aside is whether this does really improve Indian women's lot. There is some evidence suggesting that it improves financial independence it allows women to invest in dreams beyond their survival think about setting up businesses and so on and so forth so at individual level there are certainly success stories. I think the big concern though is that what seems to be

16:48happening and this is something that India's finance ministry warned us of a couple of months ago is that the emphasis on cash transfer schemes is starting to distract from policies that might more durably on some structural level improve women's lives. So we're starting to really see a crowding out of investment in the delivery of education and healthcare. It is about time that in this political class takes seriously the other half of the world's largest democracy but it would also be high time that these women were offered some more truly transformative policies. A quick word about something exciting from our sister division economist education they've got a new course developed by the kinds of senior editors and journalists that you hear

17:51on this show and it's all about geopolitics. They're joined by guests from academia, business and government to provide a nuanced view on how to think about and deal with international relations. The course is designed to give you insights you need to figure out how geopolitics affects you and your organization navigate all the uncertainty and act decisively. Sign up by May 6th and tell them we sent you. You can save 15 percent with the now familiar all caps code intelligence. Just go to education.economist.com slash intelligence. So under these lights history sharpens its pen sonnacle against Morocco the two highest rank teams on the continent summoned here for the first ever meeting. African football has promised so much

18:58for so long. One of football's greats, Pele, once predicted an African team would win the world cup by the year 2000 but it has so far failed to make a truly great impact on the world's biggest stage. John Fasmin is our senior culture correspondent and is counting down to this summer's FIFA World Cup by introducing us to 10 of the teams taking part. Among the continents record 10 teams in this year's tournament is Senegal making its third consecutive World Cup appearance and fourth overall. It is one of just four African countries who have reached the quarter finals which it did in 2002. After defeating France the defending champion they lost to Turkey on a golden goal. The Lions of Taranga as the team is known have also won the Africa Cup of Nations otherwise known as Afcon twice. Sort of. Hopes are high this year. Kulibali captains the team. He's played professionally

20:02across Europe and now Saudi Arabia. His vice captain is Idrisse Gai. Together the two have played in more than 230 international matches. Last June Senegal became the first African team to beat England in 22 attempts and in January they defeated Morocco won nothing to win that second Afcon title but that quickly turned into a farce. Senegalese players left the pitch in protest after a stoppage time penalty was awarded to Morocco in the final. Though the players did return to the field 17 minutes later the confederation of African football ruled last month that by leaving even briefly Senegal had forfeited the match and gave Morocco a win though Senegal has appealed.

21:03Off the pitch too Senegal has gone through a lot. In 2024 Basero Jumei Faye won the presidency in a landslide just two weeks after he was released from prison. Jumei as he's known promised to root out corruption create jobs for young Senegalese and loosen ties with France Senegal's former colonizer but the country's public finances remain woeful with public debt around 130% of GDP that has forced Jumei to govern more conventionally. He has raised taxes, cancelled some government contracts, cut 19 government agencies and paused some infrastructure projects. The country has still just narrowly avoided default and its bonds are junk status. Nearly 4% of Senegalese live outside of Senegal and remittances account for 10% of the country's GDP. Expect loud cheering this year for the lines of Turanga not just in Dakar but across Europe and

22:08America as well. That's all for this episode of The Intelligence. We'll see you back here tomorrow.

More episodes

More from The Intelligence from The Economist

View all episodes →